Search

This special issue features the selected works of authors who have presented papers at the 2022 iteration of the Joint Conference on Digital Libraries (JCDL) in Cologne, Germany. The motto of the conference was "Bridging Worlds" and was run as a fully hybrid event. Ten papers covering all aspects of Digital Libraries, namely Natural Language Processing, Information Retrieval, User Behavior, Scholarly Communication, Classification, Information Extraction are included in this issue. [ABSTRACT FROM AUTHOR]

This paper presents a method for developing a malware ontology structure by detecting malware instances on Twitter. The ontology represents a semi-automatic classifier fed by the data extracted from tweets. In particular, the automatic part of the presented methodology relies on a pattern-based approach to detect trigger expressions leading to new information about malware, whilst the manual one covers the evaluation of the results by domain-experts, who also validate the reliability of the semantic relationships within the ontology framework. We present preliminary results on the application of our methodology to tweets extracted from MalwareBazaar database showing how the documents’ collection analysis, through Natural Language Processing (NLP) tasks, can support the knowledge retrieval and documents’ classification procedures for building early warning system of detected malware. Results obtained from this research paper within the time framework of 2023 are referred to the previous version of the current social network X. [ABSTRACT FROM AUTHOR]

Stance detection is a crucial task in natural language processing and social computing, focusing on classifying expressed attitudes towards specific targets based on the input text. Conventional methods predominantly view stance detection as a task of target-oriented, sentence-level text classification. On popular social media platforms like Twitter, users often express their opinions through hashtags in addition to textual content within tweets. However, current methods primarily treat hashtags as data retrieval labels, neglecting to effectively utilize the semantic information they carry. In this paper, we propose a large language model knowledge-enhanced stance detection framework (LKESD) for stance detection. LKESD contains three main components: an instruction-prompted background knowledge acquisition module (IPBKA) that retrieves background knowledge of hashtags by providing handcrafted prompts to large language models (LLMs); a graph convolutional feature-enhancement module (GCFEM) is designed to extract the semantic representations of words that frequently co-occur with hashtags in the dataset by leveraging textual associations; an a knowledge fusion network (KFN) is proposed to selectively integrate graph representations and LLM features using a prompt-tuning framework. Extensive experimental results on three benchmark datasets demonstrate that our LKESD method outperforms 2.7% on all setups over compared methods, validating its effectiveness in stance detection tasks. [ABSTRACT FROM AUTHOR]

Purpose: This study aimed to investigate how the enrichment of medical documents' index terms by their comments improves the relevance and novelty of the top-ranked results retrieved by an NLP system. Design/methodology/approach: A semi-experimental pre-test and post-test research was designed to compare NLP-based indexes before and after being expanded by the comment terms. The experiments were conducted on a test collection of 13,957 documents commented by F1000-Prime reviewers. They were indexed at title, abstract, body and full-text levels. In total, 100 seed documents were randomly selected and served as queries. The textual similarity of the documents and queries was calculated using Lucene-more-like-this function and evaluated by the semantic similarity of their MeSH. The results novelty was measured using maximal marginal relevance and evaluated by their MeSH novelties. Normalized discounted cumulative gain was used to compare the basic and expanded indexes' precisions at 10, 20 and 50 top ranks. Findings: The relevance and novelty of the results ranked at the top precision points was improved after expanding the indexes by the comment terms. The finding implies that meta-texts are effective in representing their mother documents, by adding dynamic elements to their rather static contents. It also provides further evidence about the merits of the application of social intelligence and collective wisdom reflected in the actions and reactions of users in tackling the challenges faced by NLP-based systems. Originality/value: This is the first study to confirm that social comments on scientific papers improve the performance of information systems in terms of relevance and novelty. Peer review: The peer review history for this article is available at: https://publons.com/publon/10.1108/OIR-05-2022-0283. [ABSTRACT FROM AUTHOR]

Zero-Shot Learning (ZSL) aims to classify images of new categories in the testing phase without labeled images during training, using examples from categories with labeled images and some auxiliary information. The auxiliary information includes semantic attributes, textual descriptions, word embeddings, etc., for both labeled and unlabeled classes, utilizing Natural Language Processing (NLP) approaches. The word embeddings created are limited by the semantic attributes and textual descriptions where the semantics of categories are insufficient. In this paper, introduces a study for Generalized Zero-Shot Learning (GZSL), a type of ZSL, by integrating the rich semantics offered by ontology. Semantic attributes used for semantic representation are supported by ontology. Variational Autoencoder (VAE) and Generative Adversarial Network (GAN) network architectures are used together to synthesize visual features. Our work was evaluated on the AWA2 dataset, and improvement in GZSL performance was achieved. [ABSTRACT FROM AUTHOR]

The improvement of search engines for geospatial data on the World Wide Web has been a subject of research, particularly concerning the challenges in discovering and utilizing geospatial web services. Despite the establishment of standards by the Open Geospatial Consortium (OGC), the implementation of these services varies significantly among providers, leading to issues in dataset discoverability and usability. This paper presents a proof of concept for a search engine tailored to geospatial services in Switzerland. It addresses challenges such as scraping data from various OGC web service providers, enhancing metadata quality through Natural Language Processing, and optimizing search functionality and ranking methods. Semantic augmentation techniques are applied to enhance metadata completeness and quality, which are stored in a high-performance NoSQL database for efficient data retrieval. The results show improvements in dataset discoverability and search relevance, with NLP-extracted information contributing significantly to ranking accuracy. Overall, the GeoHarvester proof of concept demonstrates the feasibility of improving the discoverability and usability of geospatial web services through advanced search engine techniques. [ABSTRACT FROM AUTHOR]

Clinical abbreviation disambiguation is a crucial task in the biomedical domain, as the accurate identification of the intended meanings or expansions of abbreviations in clinical texts is vital for medical information retrieval and analysis. Existing approaches have shown promising results, but challenges such as limited instances and ambiguous interpretations persist. In this paper, we propose an approach to address these challenges and enhance the performance of clinical abbreviation disambiguation. Our objective is to leverage the power of Large Language Models (LLMs) and employ a Generative Model (GM) to augment the dataset with contextually relevant instances, enabling more accurate disambiguation across diverse clinical contexts. We integrate the contextual understanding of LLMs, represented by BlueBERT and Transformers, with data augmentation using a Generative Model, called Biomedical Generative Pre-trained Transformer (BIOGPT), that is pretrained on an extensive corpus of biomedical literature to capture the intricacies of medical terminology and context. By providing the BIOGPT with relevant medical terms and sense information, we generate diverse instances of clinical text that accurately represent the intended meanings of abbreviations. We evaluate our approach on the widely recognized CASI dataset, carefully partitioned into training, validation, and test sets. The incorporation of data augmentation with the GM improves the model's performance, particularly for senses with limited instances, effectively addressing dataset imbalance and challenges posed by similar concepts. The results demonstrate the efficacy of our proposed method, showcasing the significance of LLMs and generative techniques in clinical abbreviation disambiguation. Our model achieves a good accuracy on the test set, outperforming previous methods. [ABSTRACT FROM AUTHOR]

With the wide application of keyphrases in many Information Retrieval (IR) and Natural Language Processing (NLP) tasks, automatic keyphrase prediction has been emerging. However, these statistically important phrases are contributing increasingly less to the related tasks because the end‐to‐end learning mechanism enables models to learn the important semantic information of the text directly. Similarly, keyphrases are of little help for readers to quickly grasp the paper's main idea because the relationship between the keyphrase and the paper is not explicit to readers. Therefore, we propose to generate keyphrases with specific functions for readers to bridge the semantic gap between them and the information producers, and verify the effectiveness of the keyphrase function for assisting users' comprehension with a user experiment. A controllable keyphrase generation framework (the CKPG) that uses the keyphrase function as a control code to generate categorized keyphrases is proposed and implemented based on Transformer, BART, and T5, respectively. For the Computer Science domain, the Macro‐avgs of P@5, R@5, and F1@5 on the Paper with Code dataset are up to 0.680, 0.535, and 0.558, respectively. Our experimental results indicate the effectiveness of the CKPG models. [ABSTRACT FROM AUTHOR]

Many scientific texts are produced in Persian and available in scientific information databases through the Web. In this paper, FarsAcademic, a test collection of Persian scientific texts, has been built to implement information retrieval models among academic search comprising 102238 documents and 61 topics. While constructing FarsAcademic, we have tried to resolve the problems specific to information retrieval (IR) and natural language processing (NLP) in Persian scientific texts. Domain experts were employed to create queries within their research area, and user relevance and topical relevance were applied to improve the precision of relevance judgment of documents. Further, to improve retrieval performance in Persian scientific texts, automated query expansion was applied using one of the relevant feedback techniques, the Local Context Analysis algorithm. The result showed that query expansion techniques outperformed other information retrieval models in the Persian scientific texts retrieval task. Eventually, FarsAcademic became the only one that has been free of charge for Iranian information retrieval scholars to implement and evaluate different information retrieval models and algorithms on Persian scientific text and academic search. [ABSTRACT FROM AUTHOR]

Main path analysis is a popular method for extracting the scientific backbone from the citation network of a research domain. Existing approaches ignored the semantic relationships between the citing and cited publications, resulting in several adverse issues, in terms of coherence of main paths and coverage of significant studies. This paper advocated the semantic main path network analysis approach to alleviate these issues based on citation function analysis. A wide variety of SciBERT‐based deep learning models were designed for identifying citation functions. Semantic citation networks were built by either including important citations, for example, extension, motivation, usage and similarity, or excluding incidental citations like background and future work. Semantic main path network was built by merging the top‐K main paths extracted from various time slices of semantic citation network. In addition, a three‐way framework was proposed for the quantitative evaluation of main path analysis results. Both qualitative and quantitative analysis on three research areas of computational linguistics demonstrated that, compared to semantics‐agnostic counterparts, different types of semantic main path networks provide complementary views of scientific knowledge flows. Combining them together, we obtained a more precise and comprehensive picture of domain evolution and uncover more coherent development pathways between scientific ideas. [ABSTRACT FROM AUTHOR]

Quantifying the dissimilarity of two texts is an important aspect of a number of natural language processing tasks, including semantic information retrieval, topic classification, and document clustering. In this paper, we compared the properties and performance of different dissimilarity measures D using three different representations of texts—vocabularies, word frequency distributions, and vector embeddings—and three simple tasks—clustering texts by author, subject, and time period. Using the Project Gutenberg database, we found that the generalised Jensen–Shannon divergence applied to word frequencies performed strongly across all tasks, that D's based on vector embedding representations led to stronger performance for smaller texts, and that the optimal choice of approach was ultimately task-dependent. We also investigated, both analytically and numerically, the behaviour of the different D's when the two texts varied in length by a factor h. We demonstrated that the (natural) estimator of the Jaccard distance between vocabularies was inconsistent and computed explicitly the h-dependency of the bias of the estimator of the generalised Jensen–Shannon divergence applied to word frequencies. We also found numerically that the Jensen–Shannon divergence and embedding-based approaches were robust to changes in h, while the Jaccard distance was not. [ABSTRACT FROM AUTHOR]

Information retrieval is an important field that aims to provide a relevant document to a user information need, expressed through a query. Arabic is a challenging language that gained much attention recently in the information retrieval domain. To overcome the problems related to its complexity, many studies and techniques have been presented, most of them were conducted to solve the stemming problem. This paper presents an overview of the Arabic information retrieval process, including various text processing techniques, ranking approaches, evaluation measures, and some important information retrieval models. The paper finally presents some recent related studies and approaches in different Arabic information retrieval fields. [ABSTRACT FROM AUTHOR]

PURPOSE: Despite advances in molecular therapeutics, few anticancer agents achieve durable responses. Rational combinations using two or more anticancer drugs have the potential to achieve a synergistic effect and overcome drug resistance, enhancing antitumor efficacy. A publicly accessible biomedical literature search engine dedicated to this domain will facilitate knowledge discovery and reduce manual search and review. METHODS: We developed RetriLite, an information retrieval and extraction framework that leverages natural language processing and domain-specific knowledgebase to computationally identify highly relevant papers and extract key information. The modular architecture enables RetriLite to benefit from synergizing information retrieval and natural language processing techniques while remaining flexible to customization. We customized the application and created an informatics pipeline that strategically identifies papers that describe efficacy of using combination therapies in clinical or preclinical studies. RESULTS: In a small pilot study, RetriLite achieved an F 1 score of 0.93. A more extensive validation experiment was conducted to determine agents that have enhanced antitumor efficacy in vitro or in vivo with poly (ADP-ribose) polymerase inhibitors: 95.9% of the papers determined to be relevant by our application were true positive and the application's feature of distinguishing a clinical paper from a preclinical paper achieved an accuracy of 97.6%. Interobserver assessment was conducted, which resulted in a 100% concordance. The data derived from the informatics pipeline have also been made accessible to the public via a dedicated online search engine with an intuitive user interface. CONCLUSION: RetriLite is a framework that can be applied to establish domain-specific information retrieval and extraction systems. The extensive and high-quality metadata tags along with keyword highlighting facilitate information seekers to more effectively and efficiently discover knowledge in the combination therapy domain. [ABSTRACT FROM AUTHOR]

Over the past few years, word embeddings and bidirectional encoder representations from transformers (BERT) models have brought better solutions to learning text representations for natural language processing (NLP) and other tasks. Many NLP applications rely on pre-trained text representations, leading to the development of a number of neural network language models for various languages. However, this is not the case for Amharic, which is known to be a morphologically complex and under-resourced language. Usable pre-trained models for automatic Amharic text processing are not available. This paper presents an investigation on the essence of learned text representation for information retrieval and NLP tasks using word embeddings and BERT language models. We explored the most commonly used methods for word embeddings, including word2vec, GloVe, and fastText, as well as the BERT model. We investigated the performance of query expansion using word embeddings. We also analyzed the use of a pre-trained Amharic BERT model for masked language modeling, next sentence prediction, and text classification tasks. Amharic ad hoc information retrieval test collections that contain word-based, stem-based, and root-based text representations were used for evaluation. We conducted a detailed empirical analysis on the usability of word embeddings and BERT models on word-based, stem-based, and root-based corpora. Experimental results show that word-based query expansion and language modeling perform better than stem-based and root-based text representations, and fastText outperforms other word embeddings on word-based corpus. [ABSTRACT FROM AUTHOR]

In the era of rapid paper publications in various venues, automatic citation recommendations would be highly useful to researchers when they write papers. Local citation recommendation aims to recommend possible papers to cite given local citation contexts. Previous work mainly computes the similarity score between citation contexts and cited papers on a one-to-one basis, which is quite time-consuming. We train a pair of neural network encoders that map citation contexts and all possible cited papers to the same vector space, respectively. After that, we index the positions of all cited papers in the vector space. This makes our process for searching recommended papers considerably faster. On the other hand, existing methods tend to recommend papers that are highly similar to each other, which makes recommendations lack diversity. Therefore, we extend our algorithm to perform multi-topic recommendations. We generate multi-topic training examples based on the index we mentioned earlier. Furthermore, we specially design a multi-group contrastive learning method to train our model so that it can distinguish different topics. Empirical experiments show that our model outperforms previous methods by a wide margin. Our model is also light weighted and has been deployed online so that researchers can use it to obtain recommended citations for their own paper in real-time. • Proposed FLCR algorithm with sentence-transformer & k-d tree. • Introduced multi-topic citation recommendation for diverse contexts. • Developed large-scale dataset with 1.7 million citation contexts for evaluation. • Demonstrated significant performance improvement over previous methods. • Deployed demo for real-time citation recommendations. [ABSTRACT FROM AUTHOR]

[Display omitted] • Multi-granularity information association approaches are used to uncover multi-level geological knowledge and support accurate retrieval of mineral resource geological reports. • Extracting geological multiple feature information (topic, time, space, figure, and table) based on the multi-granularity of geological reports. • Mining potential multiple feature information associations by an improved apriori algorithm. Massive geologic report contains all kinds of multimodal geologic data information (geologic text, geologic maps, geologic tables, etc.), which contain a lot of rich geologic basic knowledge and expert experience knowledge about rocks and minerals, stratigraphic structure, geologic age, geographic location, and so on. Accurate retrieval of specific information from massive geologic data has become an important need for geologic information retrieval. However, the majority of existing research primarily revolves around extracting and associating information at a single granularity to facilitate geological semantic retrieval, which ignores many potential semantic associations, leading to ambiguity and fuzziness in semantic retrieval. To solve these problems, this paper proposes a multi-granularity (document-chapter-paragraph) geological information retrieval framework for accurate semantic retrieval. The framework firstly extracts topic feature information, spatiotemporal feature information, figure and table feature information based on the multi-granularity of geological reports. Then, an improved apriori algorithm is used to mine and visualize the associations among the feature information to discover the semantic associations of the geological reports at multiple levels of granularity. Finally, experiments are designed to validate the application of the proposed multi-granularity information retrieval framework on the accurate retrieval of geological reports. The experimental results show that the proposed multi-granularity information retrieval framework in this paper can dig deeper into underlying geo-semantic information and realize accurate retrieval. [ABSTRACT FROM AUTHOR]

PURPOSE: Despite advances in molecular therapeutics, few anticancer agents achieve durable responses. Rational combinations using two or more anticancer drugs have the potential to achieve a synergistic effect and overcome drug resistance, enhancing antitumor efficacy. A publicly accessible biomedical literature search engine dedicated to this domain will facilitate knowledge discovery and reduce manual search and review. METHODS: We developed RetriLite, an information retrieval and extraction framework that leverages natural language processing and domain-specific knowledgebase to computationally identify highly relevant papers and extract key information. The modular architecture enables RetriLite to benefit from synergizing information retrieval and natural language processing techniques while remaining flexible to customization. We customized the application and created an informatics pipeline that strategically identifies papers that describe efficacy of using combination therapies in clinical or preclinical studies. RESULTS: In a small pilot study, RetriLite achieved an F 1 score of 0.93. A more extensive validation experiment was conducted to determine agents that have enhanced antitumor efficacy in vitro or in vivo with poly (ADP-ribose) polymerase inhibitors: 95.9% of the papers determined to be relevant by our application were true positive and the application's feature of distinguishing a clinical paper from a preclinical paper achieved an accuracy of 97.6%. Interobserver assessment was conducted, which resulted in a 100% concordance. The data derived from the informatics pipeline have also been made accessible to the public via a dedicated online search engine with an intuitive user interface. CONCLUSION: RetriLite is a framework that can be applied to establish domain-specific information retrieval and extraction systems. The extensive and high-quality metadata tags along with keyword highlighting facilitate information seekers to more effectively and efficiently discover knowledge in the combination therapy domain. [ABSTRACT FROM AUTHOR]

To date, information retrieval methods in the medical field have mainly focused on English medical reports, but little work has studied Chinese electronic medical reports, especially in the field of obstetrics and gynecology. In this paper, a dataset of 180,000 complete Chinese ultrasound reports in obstetrics and gynecology was established and made publicly available. Based on the ultrasound reports in the dataset, a new information retrieval method (IKAR) is proposed to extract key information from the ultrasound reports and automatically generate the corresponding ultrasound diagnostic results. The model can both extract what is already in the report and analyze what is not in the report by inference. After applying the IKAR method to the dataset, it is proved that the method could achieve 89.38% accuracy, 91.09% recall, and 90.23% F-score. Moreover, the method achieves an F-score of over 90% on 50% of the 10 components of the report. This study provides a quality dataset for the field of electronic medical records and offers a reference for information retrieval methods in the field of obstetrics and gynecology or in other fields. [ABSTRACT FROM AUTHOR]

News feeds generate colossal amount of data consisting of important information hidden in the intricacies. State of the art methods are still at infancy in providing a very generic and publicly available solution to skim through the important information in the news from various sources and an ability to search using specific keywords in different languages. This paper focuses on designing a tool to extract semantic details from news articles published through various internet sources in various languages. The semantic information is stored within DBMS for ease of organizing and retrieving the data. Further, a querying facility to search through entire articles based on the keyword or date-based search is also proposed to view the crisp content. The news articles in English, and two Indian languages - Hindi and Malayalam are considered for experimentation. The proposed strategy consists of two main components namely, Generative model creation and Query engine. Generative model aims to extract important entities and keywords along with their relevance to the article and other similar articles using Latent Dirichlet Allocation(LDA) and Named Entity Recognition(NER). Query engine is to facilitate on the fly retrieval of semantic content from the database, based on user keyword. The search engine, along with database indexing, reduces the access time to the database thereby retrieving the information in less time. Experimental results show that the proposed method is effective in terms of quality of information and time consumed for information retrieval. [ABSTRACT FROM AUTHOR]

Since the rise of Transformer networks and large language models, cross-encoders have become the dominant architecture for various Natural Language Processing tasks. When dealing with sentence pairs, they can exploit the relationships between those pairs. On the other hand, bi-encoders can obtain a vector given a single sentence and are used in tasks such as textual similarity or information retrieval due to their low computational cost; however, their performance is inferior to that of cross-encoders. In this paper, we present Sentence-CROBI, an architecture that combines cross-encoders and bi-encoders to obtain a global representation of sentence pairs. We evaluated the proposed architecture in the paraphrase identification task using the Microsoft Research Paraphrase Corpus, the Quora Question Pairs dataset, and the PAWS-Wiki dataset. Our model obtains competitive results compared with the state-of-the-art by using model ensembles and a simple model configuration. These results demonstrate that a simple architecture that combines sentence pair and single-sentence representations without using complex pre-training or fine-tuning algorithms is a viable alternative for sentence pair tasks. [ABSTRACT FROM AUTHOR]

Semantic Textual Similarity (STS) is the task of identifying the semantic correlation between two sentences of the same or different languages. STS is an important task in natural language processing because it has many applications in different domains such as information retrieval, machine translation, plagiarism detection, document categorization, semantic search, and conversational systems. The availability of STS training and evaluation data resources for some languages such as English has led to good performance systems that achieve above 80% correlation with human judgment. Unfortunately, such required STS data resources are not available for many languages like Arabic. To overcome this challenge, this paper proposes three different approaches to generate effective STS Arabic models. The first one is based on evaluating the use of automatic machine translation for English STS data to Arabic to be used in fine-tuning. The second approach is based on the interleaving of Arabic models with English data resources. The third approach is based on fine-tuning the knowledge distillation-based models to boost their performance in Arabic using a proposed translated dataset. With very limited resources consisting of just a few hundred Arabic STS sentence pairs, we managed to achieve a score of 81% correlation, evaluated using the standard STS 2017 Arabic evaluation set. Also, we managed to extend the Arabic models to process two local dialects, Egyptian (EG) and Saudi Arabian (SA), with a correlation score of 77.5% for EG dialect and 76% for the SA dialect evaluated using dialectal conversion from the same standard STS 2017 Arabic set. [ABSTRACT FROM AUTHOR]

Efficient information search from building information models (BIMs) requires deep BIM knowledge or extensive engineering efforts for building natural language (NL)-based interfaces. To address this challenge, this paper introduces a dynamic prompt-based virtual assistant framework dubbed "BIMS-GPT" that integrates generative pre-trained transformer (GPT) technologies, supporting NL-based BIM search. To understand users' NL queries, extract relevant information from BIM databases, and deliver NL responses along with 3D visualizations, a dynamic prompt-based process was developed. In a case study, BIMS-GPT's functionality is demonstrated through a virtual assistant prototype for a hospital building. When evaluated with a BIM query dataset, the approach achieves accuracy rates of 99.5% for classifying NL queries with incorporating 2% of the data in prompts. This paper contributes to the advancement of effective and versatile virtual assistants for BIMs in the construction industry as it significantly enhances BIM accessibility while reducing the engineering and training data prerequisites for processing NL queries. • Introduced a dynamic prompt-based virtual assistant for BIM information search • Integrated BIM and GPT technologies for developing natural language interfaces • Explored prompt engineering for GPT to interpret NL queries and summarize NL answers • Improved BIM accessibility by enabling NL-based interactions with 3D visualizations • Reduced engineering and training data prerequisites for processing BIM NL queries [ABSTRACT FROM AUTHOR]

Background. The practice of traditional Chinese medicine (TCM) began several thousand years ago, and the knowledge of practitioners is recorded in paper and electronic versions of case notes, manuscripts, and books in multiple languages. Developing a method of information extraction (IE) from these sources to generate a cohesive data set would be a great contribution to the medical field. The goal of this study was to perform a systematic review of the status of IE from TCM sources over the last 10 years. Methods. We conducted a search of four literature databases for articles published from 2010 to 2021 that focused on the use of natural language processing (NLP) methods to extract information from unstructured TCM text data. Two reviewers and one adjudicator contributed to article search, article selection, data extraction, and synthesis processes. Results. We retrieved 1234 records, 49 of which met our inclusion criteria. We used the articles to (i) assess the key tasks of IE in the TCM domain, (ii) summarize the challenges to extracting information from TCM text data, and (iii) identify effective frameworks, models, and key findings of TCM IE through classification. Conclusions. Our analysis showed that IE from TCM text data has improved over the past decade. However, the extraction of TCM text still faces some challenges involving the lack of gold standard corpora, nonstandardized expressions, and multiple types of relations. In the future, IE work should be promoted by extracting more existing entities and relations, constructing gold standard data sets, and exploring IE methods based on a small amount of labeled data. Furthermore, fine-grained and interpretable IE technologies are necessary for further exploration. [ABSTRACT FROM AUTHOR]

This paper addresses the mixture symptom mention problem which appears in the structuring of Traditional Chinese Medicine (TCM). We accomplished this by disassembling mixture symptom mentions with entity relation extraction. Over 2,200 clinical notes were annotated to construct the training set. Then, an end-to-end joint learning model was established to extract the entity relations. A joint model leveraging a multihead mechanism was proposed to deal with the problem of relation overlapping. A pretrained transformer encoder was adopted to capture context information. Compared with the entity extraction pipeline, the constructed joint learning model was superior in recall, precision, and F1 measures, at 0.822, 0.825, and 0.818, respectively, 14% higher than the baseline model. The joint learning model could automatically extract features without any extra natural language processing tools. This is efficient in the disassembling of mixture symptom mentions. Furthermore, this superior performance at identifying overlapping relations could benefit the reassembling of separated symptom entities downstream. [ABSTRACT FROM AUTHOR]

Thirukkural, a Tamil classic literature, which was written in 300 BCE is a didactic literature. Though Thirukkural comprises 1330 couplets which are organized into three sections and 133 chapters, in order to retrieve meaningful Thirukkural for a given query in search systems, a better organization of the Thirukkural is needed. This paper lays such a foundation by classifying the Thirukkural into ten new categories called superclasses that is helpful for building a better Information Retrieval (IR) system. The classifier is trained using Multinomial Naïve Bayes algorithm. Each superclass is further classified into two subcategories based on the didactic information. The proposed classification framework is evaluated using precision, recall and F-score metrics and achieved an overall F-score of 82.33% and a comparison analysis has been done with the Support Vector Machine, Logistic Regression and Random Forest algorithms. An IR system is built on top of the proposed system and the performance comparison has been done with the Google search and a locally built keyword search. The proposed classification framework has achieved a mean average precision score of 89%, whereas the Google search and keyword search have yielded 59% and 68% respectively. [ABSTRACT FROM AUTHOR]

The recent development of deep learning-based natural language processing (NLP) methods has fostered many downstream applications in various fields. As one of the applications in the financial industry, fine-grained financial sentiment analysis (FSA) aims to understand the sentimental orientation, i.e., bullish or bearish, of financial texts by predicting the polarity score and has been widely applied in the financial industry stock-related opinion mining. Because of the lack of a large-scale labeled dataset and the domain-dependent nature, FSA is challenging. Previous works mainly focus on constructing and exploiting handcrafted lexicons that encode expert knowledge to enhance the semantic features in decision making, which yields improvements but are expensive to acquire. This paper proposes a lightweight regression model incorporating the statistical distribution of a term over the polarity range, say between − 1 and 1, to address the fine-grained FSA task. More concretely, we first count each word's appearance at different polarity intervals and produce a statistic-based representation for each text, which will be encoded as a corpus-level statistical feature vector by an autoencoder. Subsequently, the obtained feature vector will be integrated with the semantic feature vector in the regression model. Our experiments show such a model can produce significant improvements compared with the baseline models on two FSA subsets, i.e., news headlines and microblogs, without a computational overhead. Furthermore, we notice the signs that lexicon-based approaches have neglected can play an important role in FSA. [ABSTRACT FROM AUTHOR]

Word sense disambiguation (WSD) is an important research topic in natural language processing, which is widely applied to text classification, machine translation, and information retrieval. In order to improve disambiguation accuracy, this paper proposes a WSD method based on the graph convolutional network (GCN). Word, part of speech, and semantic category are extracted from contexts of the ambiguous word as discriminative features. Discriminative features and sentence containing the ambiguous word are used as nodes to construct the WSD graph. Word2Vec tool, Doc2Vec tool, pointwise mutual information (PMI), and TF-IDF are applied to compute embeddings of nodes and edge weights. GCN is used to fuse features of a node and its neighbors, and the softmax function is applied to determine the semantic category of the ambiguous word. Training corpus of SemEval-2007: Task #5 is adopted to optimize the proposed WSD classifier. Test corpus of SemEval-2007: Task #5 is used to test the performance of WSD classifier. Experimental results show that average accuracy of the proposed method is improved. [ABSTRACT FROM AUTHOR]

Background: Text Matching (TM) is a fundamental task of natural language processing widely used in many application systems such as information retrieval, automatic question answering, machine translation, dialogue system, reading comprehension, etc. In recent years, a large number of deep learning neural networks have been applied to TM, and have refreshed benchmarks of TM repeatedly. Among the deep learning neural networks, convolutional neural network (CNN) is one of the most popular networks, which suffers from difficulties in dealing with small samples and keeping relative structures of features. In this paper, we propose a novel deep learning architecture based on capsule network for TM, called CapsTM, where capsule network is a new type of neural network architecture proposed to address some of the short comings of CNN and shows great potential in many tasks.Methods: CapsTM is a five-layer neural network, including an input layer, a representation layer, an aggregation layer, a capsule layer and a prediction layer. In CapsTM, two pieces of text are first individually converted into sequences of embeddings and are further transformed by a highway network in the input layer. Then, Bidirectional Long Short-Term Memory (BiLSTM) is used to represent each piece of text and attention-based interaction matrix is used to represent interactive information of the two pieces of text in the representation layer. Subsequently, the two kinds of representations are fused together by BiLSTM in the aggregation layer, and are further represented with capsules (vectors) in the capsule layer. Finally, the prediction layer is a connected network used for classification. CapsTM is an extension of ESIM by adding a capsule layer before the prediction layer.Results: We construct a corpus of Chinese medical question matching, which contains 36,360 question pairs. This corpus is randomly split into three parts: a training set of 32,360 question pairs, a development set of 2000 question pairs and a test set of 2000 question pairs. On this corpus, we conduct a series of experiments to evaluate the proposed CapsTM and compare it with other state-of-the-art methods. CapsTM achieves the highest F-score of 0.8666.Conclusion: The experimental results demonstrate that CapsTM is effective for Chinese medical question matching and outperforms other state-of-the-art methods for comparison. [ABSTRACT FROM AUTHOR]

In this paper, an improved long short-term memory (LSTM)-based deep neural network structure is proposed for learning variable-length Chinese sentence semantic similarities. Siamese LSTM, a sequence-insensitive deep neural network model, has a limited ability to capture the semantics of natural language because it has difficulty explaining semantic differences based on the differences in syntactic structures or word order in a sentence. Therefore, the proposed model integrates the syntactic component features of the words in the sentence into a word vector representation layer to express the syntactic structure information of the sentence and the interdependence between words. Moreover, a relative position embedding layer is introduced into the model, and the relative position of the words in the sentence is mapped to a high-dimensional space to capture the local position information of the words. With this model, a parallel structure is used to map two sentences into the same high-dimensional space to obtain a fixed-length sentence vector representation. After aggregation, the sentence similarity is computed in the output layer. Experiments with Chinese sentences show that the model can achieve good results in the calculation of the semantic similarity. [ABSTRACT FROM AUTHOR]

Various applications in computational linguistics and artificial intelligence rely on high-performing word sense disambiguation techniques to solve challenging tasks such as information retrieval, machine translation, question answering, and document clustering. While text comprehension is intuitive for humans, machines face tremendous challenges in processing and interpreting a human's natural language. This paper presents a novel knowledge-based word sense disambiguation algorithm, namely Sequential Contextual Similarity Matrix Multiplication (SCSMM). The SCSMM algorithm combines semantic similarity, heuristic knowledge, and document context to respectively exploit the merits of local sense-based context between consecutive terms, human knowledge about terms, and a document's main topic in disambiguating terms. Unlike other algorithms, the SCSMM algorithm guarantees the capture of the maximum sentence context while maintaining the terms' order within the sentence. The proposed algorithm outperformed all other algorithms when disambiguating nouns on the combined gold standard datasets, while demonstrating comparable results to current state-of-the-art word sense disambiguation systems when dealing with each dataset separately. Furthermore, the paper discusses the impact of granularity level, ambiguity rate, sentence size, and part of speech distribution on the performance of the proposed algorithm. • Semantic similarity affects the overall performance of knowledge-based Word Sense Disambiguation (WSD) systems. • With Semantic similarity, sense heuristics, and document context, we designed a novel knowledge-based word sense disambiguation algorithm. • The Sequential Contextual Similarity Matrix Multiplication (SCSMM) algorithm captures the maximum sentence context while maintaining the words' order. • The SCSMM algorithm outperforms current WSD systems when disambiguating nouns. [ABSTRACT FROM AUTHOR]

The ubiquitous presence of surveillance systems generates massive amounts of video data. Storage and analysis of this data in real-time is a substantial challenge. There is huge potential in representing data in machine-readable and machine-interpretable format due to the presence of hidden semantics in images and videos. However, such representation requires ontology, which calls for expert domain knowledge. In this paper, a novel NLP-guided approach to generate an ontology for multimedia representation and information retrieval is proposed. A semi-automatic NLP-guided framework, which extracts all possible relations among objects is presented. This framework leverages the textual data of the domain to generate possible descriptions and actions within the domain. Relations among objects get embedded as object properties, whereas the category of an object as a class. Features and attributes of objects encode the data properties of the ontology. The proposed ontology is compared with existing multimedia ontologies and evaluated with regard to its capability to represent relations occurring in benchmark datasets, demonstrating the completeness and thorough coverage of the domain concepts. Spatial reasoning rules are established using Semantic Web Rule Language (SWRL) rules, and information retrieval is demonstrated using Description Logic (DL) and SPARQL queries. The proposed NLP-guided ontology generation approach is general enough to help in the development of ontologies for other domains as well, by providing video and textual data of the domain of interest, with limited human involvement. • Object properties extraction with textbase and NLP-guided approach. • Frequency-based semantic refinement to enhance domain-specific relations. • A novel data-driven video surveillance ontology "VizITS" for the ITS domain. • Semantic information retrieval as events, metadata, and question-answering. [ABSTRACT FROM AUTHOR]

The ubiquitous presence of surveillance systems generates massive amounts of video data. Storage and analysis of this data in real-time is a substantial challenge. There is huge potential in representing data in machine-readable and machine-interpretable format due to the presence of hidden semantics in images and videos. However, such representation requires ontology, which calls for expert domain knowledge. In this paper, a novel NLP-guided approach to generate an ontology for multimedia representation and information retrieval is proposed. A semi-automatic NLP-guided framework, which extracts all possible relations among objects is presented. This framework leverages the textual data of the domain to generate possible descriptions and actions within the domain. Relations among objects get embedded as object properties, whereas the category of an object as a class. Features and attributes of objects encode the data properties of the ontology. The proposed ontology is compared with existing multimedia ontologies and evaluated with regard to its capability to represent relations occurring in benchmark datasets, demonstrating the completeness and thorough coverage of the domain concepts. Spatial reasoning rules are established using Semantic Web Rule Language (SWRL) rules, and information retrieval is demonstrated using Description Logic (DL) and SPARQL queries. The proposed NLP-guided ontology generation approach is general enough to help in the development of ontologies for other domains as well, by providing video and textual data of the domain of interest, with limited human involvement. • Object properties extraction with textbase and NLP-guided approach. • Frequency-based semantic refinement to enhance domain-specific relations. • A novel data-driven video surveillance ontology "VizITS" for the ITS domain. • Semantic information retrieval as events, metadata, and question-answering. [ABSTRACT FROM AUTHOR]

Topic modeling is used in information retrieval to infer the hidden themes in a collection of documents and thus provides an automatic means to organize, understand and summarize large collections of textual information. Topic models also offer an interpretable representation of documents used in several downstream Natural Language Processing (NLP) tasks. Modeling techniques vary from probabilistic graphical models to the more recent neural models. This paper surveys topic models from four aspects. The first aspect categorizes different topic modeling techniques into four categories: algebraic, fuzzy, probabilistic, and neural. We review the wide variety of available models from each category, highlight differences and similarities between models and model categories using a unified perspective, investigate these models' characteristics and limitations, and discuss their proper use cases. The second aspect illustrates six criteria for proper evaluation of topic models, from modeling quality to interpretability, stability, efficiency, and beyond. Topic modeling has found applications in various disciplines, owing to its interpretability. We examine these applications along with some popular software tools which provide an implementation of some models. The fourth aspect reviews available datasets and benchmarks. Using two benchmark datasets, we conducted experiments to compare seven topic models along the proposed metrics. The discussion highlights the differences between the models and their relative suitability for various applications. It notes the relationship between evaluation metrics and proposes four key aspects to help decide which model to use for an application. Our discussion also shows that the research trends move towards developing and tuning neural topic models and leveraging the power of pre-trained language models. Finally, it highlights research gaps in developing unified benchmarks and evaluation metrics. • Categorization of topic models into algebraic, probabilistic, fuzzy and neural. • Surveys models' evaluation criteria and different metrics used. • Surveys topic modeling software and applications. • Proposes a method to help select a topic model for an application. • Compares topic models across coherence, diversity, stability and efficiency. [ABSTRACT FROM AUTHOR]

Semantic similarity measures play an important role in many natural language processing and information retrieval activities. It is highly challenging to measure semantic similarity with higher accuracy. A notable branch of semantic similarity evaluation based on information content (IC) is popular in this aspect. Intrinsic information content (IIC) models are another wing of IC based evaluation. Both IC based and IIC based approaches majorly handled similarity evaluation of nouns. Research related to semantic similarity assessment of verb pairs are rarely discussed. To bridge this gap, this work examines various IC based, IIC based approaches on verb pairs. A detailed discussion of the existing measures and their drawbacks are mentioned in this work. Strategies based on information content, length and depth of the concepts are discussed and tested on benchmark datasets. Existing intrinsic information content models are enhanced by addressing various issues like (a) dealing concepts with no path in WordNet and (b) handling the synonym sets of verb concepts. Measures based on path length, intrinsic information content, combined strategies and non-linear strategies for verb pairs are thoroughly inspected. This paper also presents novel strategies to understand novel aspects that are not addressed before. The strategies are experimented by generating the synonym sets of required parts-of-speech which proved very effective in improving the correlation with human judgment. Results on benchmark datasets specify that the proposed approaches for verb similarity will be a guiding factor for understanding the natural language processing tasks. • Semantic similarity measures to address the verb pairs. • Path-based, information content based, combined and non-linear strategies for evaluating semantic similarity. • Intrinsic information content models for evaluating semantic similarity. • In this paper, handling the word pairs which does not have path in the semantic network proved effective. • Generating proper synonym sets of the concepts, improved the correlation of the strategies proposed. [ABSTRACT FROM AUTHOR]

• Propose a hybrid Transformer approach for Chinese NER. • Take the latent words as the additional features to improve NER task. • Find an efficient strategy to fine-tune BERT with other modules. • Apply relative position encodings to improve Transformer. Named entity recognition (NER) plays an important role in many downstream tasks of natural language processing, such as knowledge extraction and information retrieval. NER of Chinese is more challenging than that of English due to lack of the explicit word boundary. Features augmentation is a potential way to improve NER model of Chinese. Pre-trained models can implicitly preserve prior knowledge with additional features. This paper proposes a hybrid Transformer approach, which first utilize the fused additional features embeddings (e.g. char embeddings, bigram embeddings, lattice embeddings and BERT embeddings) as distributed representations to augment the representation ability of model. In addition, a new training strategy named DF strategy is proposed to efficiently fine-tune Bidirectional Encoder Representations from Transformers (BERT) and other embeddings in balance. Then, the proposed model can perceive the relations of features by introducing relative position embeddings to an additional adapted Transformer encoder. Lastly, a standard Conditional Random Field is used to alleviate the obvious tag errors. The proposed model is applied to four representative Chinese datasets to investigate its performance. Experiments results show that the proposed model outperforms the other popular models in terms of accuracy. The proposed BL-BTC model can effectively improve the recognition performance of formal and informal texts. [ABSTRACT FROM AUTHOR]

As e-commerce markets have gradually expanded, online shopping malls have provided various services aiming to secure competitiveness. A service for providing an accurate and prompt response when a customer writes an inquiry regarding a product represents a space directly connected to the customer and plays an important role, as it is directly related to product sales. However, the current online shopping mall answering service has disadvantages, e.g., it takes time for an administrator to write an answer directly, or to provide an answer within a set of answers. In this paper, we propose an answer framework for solving this problem, based on customer reviews. When a user writes a query, the framework provides an appropriate answer in real time through the system's question-and-answer pairs and customer reviews. The framework's performance is verified through a qualitative evaluation. In addition, it is confirmed that a customized model for reflecting the characteristics of each shopping mall can be created by using additional information from the collected data. The proposed framework is expected to support customers' online shopping through more reliable and efficient information retrieval, and to reduce shopping mall operation and maintenance costs. • We proposed an answer framework using Q&A pairs and customer reviews. • We obtained high performance in customer reviews category classification. • We provided customer reviews suitable for users' inquiries along with Q&A data. • A qualitative evaluation was conducted on the system output for any input query. • We provided a customized model framework for the Q&A system. [ABSTRACT FROM AUTHOR]

Web search is widely used to find online medical advice. As such, health-related information access requires retrieval algorithms capable of promoting reliable documents and filtering out unreliable ones. To this end, different types of components, such as query-document matching features, passage relevance estimation and AI-based reliability estimators, need to be combined. In this paper, we propose an entire pipeline for misinformation detection, based on the fusion of multiple content-based features. We present experiments which study the influence of each pipeline stage for the target task. Our technological solution incorporates signals from technologies derived from diverse research fields, including search, deep learning for natural language processing, as well as advanced supervised and unsupervised learning. To combine evidence, different score fusion strategies are compared, including unsupervised rank fusion techniques and learning-to-rank methods. The reference framework for empirically validating our solution is the TREC Health Misinformation Track, which provides several challenging subtasks that foster research on the identification of reliable and correct information for health-related decision making tasks. More specifically, we address a total recall task, the goal of which is to identify all the documents conveying incorrect information for a specific set of topics, and an ad-hoc retrieval task, aiming to rank credible and correct information over incorrect information. All variants are evaluated with an assorted set of effectiveness metrics, which includes standard search measures, such as R-Precision, Average Precision or Normalised Discounted Cumulative Gain, and innovative metrics based on the compatibility between the ranked output and two reference rankings composed of helpful and harmful documents, respectively. Our experiments demonstrate the effectiveness of the proposed pipeline stages and indicate that sophisticated supervised fusion methods do not fare better than simpler fusion alternatives. Additionally, for reliability estimation, unsupervised textual similarity performs better than textual classification based on supervised learning. The results also show that the presented approach is highly competitive when compared with state-of-the-art solutions for the same problem. • A complete system for health misinformation detection is proposed. • Several content-based features are proposed for identifying health misinformation. • The performance of the different features and against external baselines is reported. • New metrics considering harmfulness and helpfulness of retrieved documents are used. • The technology developed is freely available for other researchers or stakeholders. [ABSTRACT FROM AUTHOR]

The rapid proliferation of text data has lead to an increase in the use of Information Extraction (IE) techniques to automatically extract key information in a fast and effective manner. Relation Extraction (RE), a sub-task of IE focuses on extracting semantic relations from free natural language text and is crucial for further applications including Question Answering, Information Retrieval, Knowledge Base construction, Text Summarization, etc. Literature shows that supervised learning approaches were widely used in RE. However, the performance of supervised methodologies depend on the availability of domain-specific annotated datasets which is not viable for many of the domains including legal, financial, insurance etc. In recent times, Open Information Extraction (OIE) techniques address this issue, by facilitating domain-independent extraction of relations from large text corpora with no demand for domain-specific tagged data and predefined relation classes. Even though OIE systems are fast and simple to implement, they are less effective in handling complex sentences, and often produce redundant extractions. This paper proposes an efficient RE system to extract domain-specific relations from natural language text, consisting of Knowledge-based and Semi-supervised learning systems, integrated with domain ontology. We evaluated the performance of proposed work on 'judicial domain" as a use case and found that it overcomes the flaws and limitations of existing RE approaches, by achieving better results in terms of precision and recall. On further analysis, we found that the proposed system outperforms existing cutting-edge OIE systems on varying sentence length and complexity. • Automatic extraction of semantic relations from large volume of natural language text belonging to closed domain, particularly judicial domain. • Combination of knowledge based and semi-supervised learning systems to extract domain-specific relations from complex text that lacks domain-specific labelled data. • Solution to various challenges in handling complex sentences in passive voice and containing conjunctive forms. • Evaluation of state-of-the-art open information extraction approaches on judicial text with varying complexity and length. • First attempt to provide a solution for extracting domain-specific relations from judicial text. [ABSTRACT FROM AUTHOR]

Modern cross-lingual document retrieval models are capable of finding documents relevant to the query. However, they do not have the capabilities for explaining why the document is relevant. This paper proposes a novel learning-to-rank model named LM-EMD that uses the multilingual BERT language model and Earth Mover's Distance (EMD) to measure the document's relevancy to the input query and provide interpretable insights into why a document is relevant. The model uses the query and document token's contextual embeddings generated with multilingual BERT to measure their distances in the embedding space, which are then used by EMD to calculate the document's relevance score and identify which document tokens contribute the most to its relevancy. We evaluate the model on five language pairs of varying degrees of similarity and analyze its performance. We find that the model (1) performs similar as the best performing comparing model on high-resource languages, (2) is less effective on low-resource languages, and (3) provides insight into why a document is relevant to the query. • Theoretical background on using optimal transport for measuring document relevancy. • A novel learning-to-rank model outputting interpretable document relevance scores. • The model performs best on high-resource language pair data sets. • Performs an analysis of the impact the loss function has on the model's performance. [ABSTRACT FROM AUTHOR]