Search

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]

Biomedical question answering (QA) represents a growing concern among industry and academia due to the crucial impact of biomedical information. When mapping and ranking candidate snippet answers within relevant literature, current QA systems typically refer to information retrieval (IR) techniques: specifically, query processing approaches and ranking models. However, these IR-based approaches are insufficient to consider both syntactic and semantic relatedness and thus cannot formulate accurate natural language answers. Recently, deep learning approaches have become well-known for learning optimal semantic feature representations in natural language processing tasks. In this paper, we present a deep ranking recursive autoencoders (rankingRAE) architecture for ranking question-candidate snippet answer pairs (Q-S) to obtain the most relevant candidate answers for biomedical questions extracted from the potentially relevant documents. In particular, we convert the task of ranking candidate answers to several simultaneous binary classification tasks for determining whether a question and a candidate answer are relevant. The compositional words and their random initialized vectors of concatenated Q-S pairs are fed into recursive autoencoders to learn the optimal semantic representations in an unsupervised way, and their semantic relatedness is classified through supervised learning. Unlike several existing methods to directly choose the top-K candidates with highest probabilities, we take the influence of different ranking results into consideration. Consequently, we define a listwise "ranking error" for loss function computation to penalize inappropriate answer ranking for each question and to eliminate their influence. The proposed architecture is evaluated with respect to the BioASQ 2013-2018 Six-year Biomedical Question Answering benchmarks. Compared with classical IR models, other deep representation models, as well as some state-of-the-art systems for these tasks, the experimental results demonstrate the robustness and effectiveness of rankingRAE. [ABSTRACT FROM AUTHOR]

More than ever, technical inventions are the symbol of our society’s advance. Patents guarantee their creators protection against infringement. For an invention being patentable, its novelty and inventiveness have to be assessed. Therefore, a search for published work that describes similar inventions to a given patent application needs to be performed. Currently, this so-called search for prior art is executed with semi-automatically composed keyword queries, which is not only time consuming, but also prone to errors. In particular, errors may systematically arise by the fact that different keywords for the same technical concepts may exist across disciplines. In this paper, a novel approach is proposed, where the full text of a given patent application is compared to existing patents using machine learning and natural language processing techniques to automatically detect inventions that are similar to the one described in the submitted document. Various state-of-the-art approaches for feature extraction and document comparison are evaluated. In addition to that, the quality of the current search process is assessed based on ratings of a domain expert. The evaluation results show that our automated approach, besides accelerating the search process, also improves the search results for prior art with respect to their quality. [ABSTRACT FROM AUTHOR]

Web of Things (WoT) platforms are growing fast so as the needs for composing WoT apps more easily and efficiently. We have recently commenced the campaign to develop an interface where users can issue requests for WoT apps entirely in natural language. This requires an effort to build a system that can learn to identify relevant WoT functions that fulfill user’s requests. In our preceding work, we trained a supervised learning system with thousands of publicly-available IFTTT app recipes based on conditional random fields (CRF). However, the sub-par accuracy and excessive training time motivated us to devise a better approach. In this paper, we present a novel solution that creates a separate learning engine for each trigger service. With this approach, parallel and incremental learning becomes possible. For inference, our system first identifies the most relevant trigger service for a given user request by using an information retrieval technique. Then, the learning engine associated with the trigger service predicts the most likely pair of trigger and action functions. We expect that such two-phase inference method given parallel learning engines would improve the accuracy of identifying related WoT functions. We verify our new solution through the empirical evaluation with training and test sets sampled from a pool of refined IFTTT app recipes. We also meticulously analyze the characteristics of the recipes to find future research directions. [ABSTRACT FROM AUTHOR]