Search

Question Answering (QA) systems based on Information Retrieval return precise answers to natural language questions, extracting relevant sentences from document collections. However, questions and sentences cannot be aligned terminologically, generating errors in the sentence retrieval. In order to augment the effectiveness in retrieving relevant sentences from documents, this paper proposes a hybrid Query Expansion (QE) approach, based on lexical resources and word embeddings , for QA systems. In detail, synonyms and hypernyms of relevant terms occurring in the question are first extracted from MultiWordNet and, then, contextualized to the document collection used in the QA system. Finally, the resulting set is ranked and filtered on the basis of wording and sense of the question, by employing a semantic similarity metric built on the top of a Word2Vec model. This latter is locally trained on an extended corpus pertaining the same topic of the documents used in the QA system. This QE approach is implemented into an existing QA system and experimentally evaluated, with respect to different possible configurations and selected baselines, for the Italian language and in the Cultural Heritage domain, assessing its effectiveness in retrieving sentences containing proper answers to questions belonging to four different categories.

Semantic web image retrieval is useful to end-users for semantic image searches over the Internet. This paper aims to develop image retrieval techniques for large-scale web image databases. An advanced retrieval system, termed Multi-concept Retrieval using Bimodal Deep Learning (MRBDL), is proposed and implemented using Convolutional Neural Networks (CNNs) which can effectively capture semantic correlations between a visual image and its free contextual tags. Different from existing approaches using multiple and independent concepts in a query, MRBDL considers multiple concepts as a holistic scene for retrieval model learning. In particular, we first use a bimodal CNN to train a holistic scene classifier in two modalities, and then semantic correlations of the sub-concepts included in the images are leveraged to boost holistic scene recognition. The predicted semantic scores obtained from holistic scene classifier are combined with complementary information on web images to improve the retrieval performance. Experiments have been carried out over two publicly available web image databases. The results show that our proposed approach performs favorably compared with several other state-of-the-art methods .

This paper presents the application of a Wikipedia-based bag of concepts (WikiBoC) document representation to cross-language text classification (CLTC). Its main objective is to alleviate the major drawbacks of the state-of-the-art CLTC approaches typically based on the machine translation (MT) of documents, which are represented as bags of words (BoW). We propose a technique called cross-language concept matching (CLCM), to convert concept-based representations of documents from one language to another using Wikipedia correspondences between concepts in different languages and thus not relying on automated full-text translations. We describe two proposals: the first proposal consists in the use of the WikiBoC representation in conjunction with the CLCM technique (WikiBoC-CLCM) to classify documents written in a language L1 by using a SVM algorithm that was trained with documents written in another language L2; the second proposal consists of a hybrid model for representing documents that combines WikiBoC-CLCM with the classic BoW-MT approach. To evaluate the two proposals we conducted several experiments with three cross-lingual corpora: the JRC-Acquis corpus and two purpose-built corpora composed of Wikipedia articles. The first proposal outperforms state-of-the-art approaches when training sequences are short, achieving performance increases up to 233.33%. The second proposal outperforms state-of-the-art approaches in the whole range of training sequences, achieving performance increases up to 23.78%. Results obtained show the benefits of the WikiBoC-CLCM approach, since concepts extracted from documents add useful information to the classifier, thus improving its performance.

A framework to generate pseudo-documents suitable for topic modeling is proposed.An instantiation of the framework based on word vector representations is presented.Results of NPMI and F1 obtained are better than those of state-of-the art methods. Short texts are everywhere in the Web, including messages posted in social media, status messages and blog comments, and uncovering the topics of this type of messages is crucial to a wide range of applications, e.g., context analysis and user characterization. Extracting topics from short text is challenging because of the dependence of conventional methods, such as Latent Dirichlet Allocation, in words co-occurrence, which in short text is rare and make these methods suffer from severe data sparsity. This paper proposes a general framework for topic modeling of short text by creating larger pseudo-document representations from the original documents. In the framework, document components (e.g., words or bigrams) are defined over a metric space, which provides information about the similarity between them. We present two simple, effective and efficient methods that specialize our general framework to create larger pseudo-documents. While the first method considers word co-occurrence to define the metric space, the second relies on distributed word vector representations. The pseudo-documents generated can be given as input to any topic modeling algorithm. Experiments run in seven datasets and compared against state-of-the-art methods for extracting topics by generating pseudo-documents or modifying current topic modeling methods for short text show the methods significantly improve results in terms of normalized pointwise mutual information. A classification task was also used to evaluate the quality of the topics in terms of document representation, where improvements in F1 varied from 1.5 to 15%.

Social networks messaging typically contains a lot of implicit linguistic information partially due to restrictions on a message's length (i.e., few named entities, short sentences, no discourse structure, etc.). This may significantly impact several applications including opinion mining, sentiment analysis, etc., as data collection tasks such as opinion retrieval tasks will fail to obtain all the relevant messages whenever the target topic, objects, or features are not explicit within the texts. In order to address these issues, in this paper a novel adaptive approach for opinion retrieval is proposed. It combines natural-language co-referencing techniques, features-based linguistic preprocessing and memory-based learning to resolving implicit co-referencing within informal opinion texts by using underlying hierarchies of thread messages. Experiments were conducted to assess the ability of the model to improve opinion retrieval by resolving implicit entities and features, showing the promise of our opinion retrieval approach when compared to state-of-the-art methods using text data from social networks.