Semantic Similarity Model Research Articles

Developing a Cross-lingual Semantic Word Similarity Corpus for English–Urdu Language Pair

Semantic word similarity is a quantitative measure of how much two words are contextually similar. Evaluation of semantic word similarity models requires a benchmark corpus. However, despite the millions of speakers and the large digital text of the Urdu language on the Internet, there is a lack of benchmark corpus for the Cross-lingual Semantic Word Similarity task for the Urdu language. This article reports our efforts in developing such a corpus. The newly developed corpus is based on the SemEval-2017 task 2 English dataset, and it contains 1,945 cross-lingual English–Urdu word pairs. For each of these pairs of words, semantic similarity scores were assigned by 11 native Urdu speakers. In addition to corpus generation, this article also reports the evaluation results of a baseline approach, namely “Translation Plus Monolingual Analysis” for automated identification of semantic similarity between English–Urdu word pairs. The results showed that the path length similarity measure performs better for the Google and Bing translated words. The newly created corpus and evaluation results are freely available online for further research and development.

Discovery and Validation of Key Biomarkers Based on Immune Infiltrates in Alzheimer's Disease.

BackgroundAs the most common neurodegenerative disease, Alzheimer’s disease (AD) leads to progressive loss of cognition and memory. Presently, the underlying pathogenic genes of AD patients remain elusive, and effective disease-modifying therapy is not available. This study explored novel biomarkers that can affect diagnosis and treatment in AD based on immune infiltration.MethodsThe gene expression profiles of 139 AD cases and 134 normal controls were obtained from the NCBI GEO public database. We applied the computational method CIBERSORT to bulk gene expression profiles of AD to quantify 22 subsets of immune cells. Besides, based on the use of the Least Absolute Shrinkage Selection Operator (LASSO), this study also applied SVM-RFE analysis to screen key genes. GO-based semantic similarity and logistic regression model analyses were applied to explore hub genes further.ResultsThere was a remarkable significance in the infiltration of immune cells between the subgroups. The proportions for monocytes, M0 macrophages, and dendritic cells in the AD group were significantly higher than those in the normal group, while the proportion of some cells was lower than that of the normal group, such as NK cell resting, T-cell CD4 naive, T-cell CD4 memory activation, and eosinophils. Additionally, seven genes (ABCA2, CREBRF, CD72, CETN2, KCNG1, NDUFA2, and RPL36AL) were identified as hub genes. Then we performed the analysis of immune factor correlation, gene set enrichment analysis (GSEA), and GO based on seven hub genes. The AUC of ROC prediction model in test and validation sets were 0.845 and 0.839, respectively. Eventually, the mRNA expression analysis of ABCA2, NDUFA2, CREBRF, and CD72 revealed significant differences among the seven hub genes and then was confirmed by RT-PCR.ConclusionA model based on immune cell infiltration might be used to forecast AD patients’ diagnosis, and it provided a new perspective for AD treatment targets.

A Semantic Similarity Model to Compare Heterogeneous Data Sources to Augment Engineering Data with New Failure modes in Automotive Industry

In real life industry, in-time exposure of symptoms and their failure modes observed during fault events provide key signals to system engineers to take necessary corrective actions for effectively arresting product defects. A real-life ontology-based semantic similarity system is employed for automatic comparison of engineering data (comes in the form of design failure mode effect analysis, DFMEA data) with the field repair data collected during product warranty period. Given the complexity of engineering data and the overwhelming volume (hundreds of millions of data points) of field repair data makes identification of new symptoms and failure modes from the first principles an impractical task. Typically, the engineering data is recorded by using technical vocabulary, e.g. unstable electric contact, Seat Belt comfort per MVSS208, whereas the field repair data is highly unstructured in nature. Consequently, we observe following types of noises in the field repair data – abbreviated text entries, inconsistent use of vocabulary (‘seat buckle is damaged’ vs ‘buckle unlatching’), and finally the incomplete text entries. More importantly, limited mental mapping capacity of a human agent limits discovery of new symptoms and failure modes from the industrial scale data. Not surprisingly, the text mining and semantic similarity are gaining a serious attention due to their ability to automatically discover the knowledge assets buried in unstructured text by training machines to compare and link high volume of data. In our approach, initially the key constructs (e.g. symptoms, failure modes) from the data are annotated by using the domain ontology. These constructs are then used to construct pairs of terms and pairs of tuples, which are used to compute pair-to-pair and tuple-to-tuple semantic similarity respectively. Finally, the text-to-text semantic similarity is calculated by combining other two semantic similarity scores. It is used to determine whether new symptoms or failure modes from the field repair data can be used to augment the DFMEA data. The proposed method is implemented as a prototype tool and its performance is validated by using real-life data from automobile domain. On an average, our system has F1 score of 0.75 and 0.78 in discovering and identifying new symptoms and synonym symptoms respectively, whereas it achieved the F1 score of 0.72 and 0.68 in discovering new failure modes and in identifying synonym failure modes respectively. The fault detection rate is improved by 35%, whereas the fault isolation rate is improved by 40.5%.

An OGC web service geospatial data semantic similarity model for improving geospatial service discovery

Abstract Open Geospatial Consortium (OGC) Web Services (OWS) are highly significant for geospatial data sharing and widely used in many scientific fields. However, those services are hard to find and utilize effectively. Focusing on addressing the big challenge of OWS resource discovery, we propose a measurement model that integrates spatiotemporal similarity and thematic similarity based on ontology semantics to generate a more efficient search method: OWS Geospatial Data Semantic Similarity Model (OGDSSM)-based search engine for semantically enabled geospatial data service discovery that takes into account the hierarchy difference of geospatial service documents and the number of map layers. We implemented the proposed OGDSSM-based semantic search algorithm on United States Geological Survey mineral resources geospatial service discovery. The results show that the proposed search method has better performance than the existing search engines that are based on keyword-based matching, such as Lucene, when recall, precision, and F-measure are taken into consideration. Furthermore, the returned results are ranked based on semantic similarity, which makes it easier for users to find the most similar geospatial data services. Our proposed method can thus enhance the performance of geospatial data service discovery for a wide range of geoscience applications.

Using Character-Level and Entity-Level Representations to Enhance Bidirectional Encoder Representation From Transformers-Based Clinical Semantic Textual Similarity Model: ClinicalSTS Modeling Study.

BackgroundWith the popularity of electronic health records (EHRs), the quality of health care has been improved. However, there are also some problems caused by EHRs, such as the growing use of copy-and-paste and templates, resulting in EHRs of low quality in content. In order to minimize data redundancy in different documents, Harvard Medical School and Mayo Clinic organized a national natural language processing (NLP) clinical challenge (n2c2) on clinical semantic textual similarity (ClinicalSTS) in 2019. The task of this challenge is to compute the semantic similarity among clinical text snippets.ObjectiveIn this study, we aim to investigate novel methods to model ClinicalSTS and analyze the results.MethodsWe propose a semantically enhanced text matching model for the 2019 n2c2/Open Health NLP (OHNLP) challenge on ClinicalSTS. The model includes 3 representation modules to encode clinical text snippet pairs at different levels: (1) character-level representation module based on convolutional neural network (CNN) to tackle the out-of-vocabulary problem in NLP; (2) sentence-level representation module that adopts a pretrained language model bidirectional encoder representation from transformers (BERT) to encode clinical text snippet pairs; and (3) entity-level representation module to model clinical entity information in clinical text snippets. In the case of entity-level representation, we compare 2 methods. One encodes entities by the entity-type label sequence corresponding to text snippet (called entity I), whereas the other encodes entities by their representation in MeSH, a knowledge graph in the medical domain (called entity II).ResultsWe conduct experiments on the ClinicalSTS corpus of the 2019 n2c2/OHNLP challenge for model performance evaluation. The model only using BERT for text snippet pair encoding achieved a Pearson correlation coefficient (PCC) of 0.848. When character-level representation and entity-level representation are individually added into our model, the PCC increased to 0.857 and 0.854 (entity I)/0.859 (entity II), respectively. When both character-level representation and entity-level representation are added into our model, the PCC further increased to 0.861 (entity I) and 0.868 (entity II).ConclusionsExperimental results show that both character-level information and entity-level information can effectively enhance the BERT-based STS model.

Semantic Technologies for Semantic Applications. Part 2. Models of Comparative Text Semantics

Both parts of this paper discuss the basic aspects of semantic computing, semantic technologies, and semantic applications applied to NL-text big data processing for knowledge extracting and decision making. The basic components of the corresponding systems and technologies are reviewed, which include ontologies and semantic models of their use, semantic resources, and semantic component. The semantic resources contain knowledge about the semantics and means for refinement of this semantics. The semantic component of the technology is used to formally describe the meaning of NL-entities and numerically evaluate their pairwise semantic similarity. The main focus of this part is on numerical models of pairwise semantic similarity of NL-entities. These models are important for solving tasks of text semantic clustering and classification and their various applications. Various types of semantic relatedness and semantic similarity measures for NL-entities in the context of semantic computing tasks are discussed and compared. Problems that constrain the practical use of semantic technologies for the development of semantic applications are analyzed.

Chatbot Designing Information Service for New Student Registration Based on AIML and Machine Learning

One of the efforts made by universities to serve prospective students is by providing consulting services and information that is usually carried out directly at the booth provided, through phone service or live chat support available on the college website. Increased visitors will result in waiting times due to limited availability of officers, which results in decreased satisfaction of prospective new students, moreover this service is only available during campus operating hours. One alternative solution to overcome this problem is to use Chatbot, able to answer questions raised by prospective new students which can be categorized as Frequently Asked Questions abbreviated as FAQ. Chatbot technology can be developed with a variety of AI (Artificial Intelligence) techniques. One of them is the AIML (Artificial Intelligence Markup Language) technique. One of the main drawbacks of AIML is that there is no reasoning ability so a learning system that is focused on supervised learning is needed. In the chatbot that will be built the learning process uses a selective neural conversational model or commonly called the Deep Semantic Similarity Model (DSSM) developed by Microsoft. Meanwhile, the measurement of chatbot performance will be done using Confusion Matrix which is a method of evaluating the performance of the algorithm from Machine Learning (ML). The results of the study stated that the chatbot system that was built was able to answer questions posed by prospective students properly and correctly while the questions were available in the chatbot knowledge base.

Research on Distributed Search Technology of Multiple Data Sources Intelligent Information Based on Knowledge Graph

The traditional information search technology performs full-text indexing on the data in the Internet, searches for information by means of keyword matching index, and returns information to the user. This retrieval method does not accurately understand the user’s needs, and returns relevant links rather than the information the user needs. The user needs to find relevant information from the linked documents. In order to improve the shortcomings of the above traditional search technology, this paper is based on the knowledge map of multi-data source intelligent information distributed search technology, through data acquisition in the Internet, complete the transformation of data to knowledge to form a knowledge network and provide information retrieval. This paper studies the construction of knowledge maps for application domain representation, proposes a semantic similarity model with constraints and implicit feedback correction mechanism, and explores the realization of intelligent information search under certain conditions. Through the analysis of prototype experimental data in the application field, the accuracy of information search based on knowledge map can reach 90%, which has strong practicability.

Understanding the spatial dimension of natural language by measuring the spatial semantic similarity of words through a scalable geospatial context window

Measuring the semantic similarity between words is important for natural language processing tasks. The traditional models of semantic similarity perform well in most cases, but when dealing with words that involve geographical context, spatial semantics of implied spatial information are rarely preserved. Geographic information retrieval (GIR) methods have focused on this issue; however, they sometimes fail to solve the problem because the spatial and textual similarities of words are considered and calculated separately. In this paper, from the perspective of spatial context, we consider the two parts as a whole—spatial context semantics, and we propose a method that measures spatial semantic similarity using a sliding geospatial context window for geo-tagged words. The proposed method was first validated with a set of simulated data and then applied to a real-world dataset from Flickr. As a result, a spatial semantic similarity model at different scales is presented. We believe this model is a necessary supplement for traditional textual-language semantic analyses of words obtained by word-embedding technologies. This study has the potential to improve the quality of recommendation systems by considering relevant spatial context semantics, and benefits linguistic semantic research by emphasising the spatial cognition among words.

Collective Entity Disambiguation Based on Hierarchical Semantic Similarity

Entity disambiguation involves mapping mentions in texts to the corresponding entities in a given knowledge base. Most previous approaches were based on handcrafted features and failed to capture semantic information over multiple granularities. For accurately disambiguating entities, various information aspects of mentions and entities should be used in. This article proposes a hierarchical semantic similarity model to find important clues related to mentions and entities based on multiple sources of information, such as contexts of the mentions, entity descriptions and categories. This model can effectively measure the semantic matching between mentions and target entities. Global features are also added, including prior popularity and global coherence, to improve the performance. In order to verify the effect of hierarchical semantic similarity model combined with global features, named HSSMGF, experiments were carried out on five publicly available benchmark datasets. Results demonstrate the proposed method is very effective in the case that documents have more mentions.

Semantic similarity measurement: an intrinsic information content model

Ontology dependent Semantic Similarity (SS) measurement has emerged as a new research paradigm in finding the semantic strength between any two entities. In this regard, as observed, the information theoretic intrinsic approach yields better accuracy in correlation with human cognition. The precision of such a technique highly depends on how accurately we calculate Information Content (IC) of concepts and its compatibility with a SS model. In this work, we develop an intrinsic IC model to facilitate better SS measurement. The proposed model has been evaluated using three vocabularies, namely SNOMED CT, MeSH and WordNet against a set of benchmark data sets. We compare the results with the state-of-the-art IC models. The results show that the proposed intrinsic IC model yields a high correlation with human assessment. The article also evaluates the compatibility of the proposed IC model and the other existing IC models in combination with a set of state-of-the-art SS models.

An Approach for Radicalization Detection Based on Emotion Signals and Semantic Similarity

The Internet has become an important tool for modern terrorist groups as a means of spreading their propaganda messages and recruitment purposes. Previous studies have shown that the analysis of social signs can help in the analysis, detection, and prediction of radical users. In this work, we focus on the analysis of affect signs in social media and social networks, which has not been yet previously addressed. The article contributions are: (i) a novel dataset to be used in radicalization detection works, (ii) a method for utilizing an emotion lexicon for radicalization detection, and (iii) an application to the radical detection domain of an embedding-based semantic similarity model. Results show that emotion can be a reliable indicator of radicalization, as well as that the proposed feature extraction methods can yield high-performance scores.

Car FAQ Assistant Based on BILSTM-Siamese Network

With the development of artificial intelligence, the automatic question answering technology has been paid more and more attention. Along with the technology’s maturing, the problem is gradually exposed. This technique has several major difficulties. There is semantic recognition. How accurate are the answers to the questions? Because Chinese grammar is relatively complex. There are many different ways of saying the same thing. There is a powerful challenge to semantic recognition. Due to poor semantic recognition, the accuracy of the corresponding questions and answers is low. Aiming at this kind of questions, we build a Q & A library for common use of cars. Firstly, TFIDF method is used to build a basic FAQ retrieval system. Then we use BILSTM-siamese network to construct a semantic similarity model. The accuracy rate on the test set was 99.52%. The final FAQ system: first retrieves 30 most similar questions using the TFIDF model, then uses BILSTM-siamese network matching and returns the answer of the most similar question. The system has a lot to be improved. For example, how to combine this system with speech recognition? How to realize speech recognition will be a great challenge.

Improved scalability in mining using ontology record linkage algorithm

Record linkage offers wide role in record identification and relevant datasets matching. The conventional researchers use probabilistic approach to identify reliable and unique datasets. Record linkage with probabilistic approach exploits data, which are common to an individual record pair. Classical methods have equality based record linkage in common fields. Therefore, errors associated with record linkage reduce the scalability. In this paper, a similarity between individual values of record pairs is improved using ontology-based semantic similarity model. Semantic similarity between the records is tested successfully using angle based neighborhood graph. To validate the proposed approach, a conventional record linkage algorithm is compared with angle based neighborhood ontology record linkage technique, which achieves improved accuracy and scalability. Finally, the accuracy of identifying similar semantic matches is more scalable in proposed technique than conventional methods.

Speech understanding for spoken dialogue systems: From corpus harvesting to grammar rule induction

We investigate algorithms and tools for the semi-automatic authoring of grammars for spoken dialogue systems (SDS) proposing a framework that spans from corpora creation to grammar induction algorithms. A realistic human-in-the-loop approach is followed balancing automation and human intervention to optimize cost to performance ratio for grammar development. Web harvesting is the main approach investigated for eliciting spoken dialogue textual data, while crowdsourcing is also proposed as an alternative method. Several techniques are presented for constructing web queries and filtering the acquired corpora. We also investigate how the harvested corpora can be used for the automatic and semi-automatic (human-in-the-loop) induction of grammar rules. SDS grammar rules and induction algorithms are grouped into two types, namely, low- and high-level. Two families of algorithms are investigated for rule induction: one based on semantic similarity and distributional semantic models, and the other using more traditional statistical modeling approaches (e.g., slot-filling algorithms using Conditional Random Fields). Evaluation results are presented for two domains and languages. High-level induction precision scores up to 60% are obtained. Results advocate the portability of the proposed features and algorithms across languages and domains.

A hybrid similarity measure method for patent portfolio analysis

Abstract Similarity measures are fundamental tools for identifying relationships within or across patent portfolios. Many bibliometric indicators are used to determine similarity measures; for example, bibliographic coupling, citation and co-citation, and co-word distribution. This paper aims to construct a hybrid similarity measure method based on multiple indicators to analyze patent portfolios. Two models are proposed: categorical similarity and semantic similarity. The categorical similarity model emphasizes international patent classifications (IPCs), while the semantic similarity model emphasizes textual elements. We introduce fuzzy set routines to translate the rough technical (sub-) categories of IPCs into defined numeric values, and we calculate the categorical similarities between patent portfolios using membership grade vectors. In parallel, we identify and highlight core terms in a 3-level tree structure and compute the semantic similarities by comparing the tree-based structures. A weighting model is designed to consider: 1) the bias that exists between the categorical and semantic similarities, and 2) the weighting or integrating strategy for a hybrid method. A case study to measure the technological similarities between selected firms in China’s medical device industry is used to demonstrate the reliability our method, and the results indicate the practical meaning of our method in a broad range of informetric applications.

QSem: A novel question representation framework for question matching over accumulated question–answer data

This paper proposes a novel question representation framework to assist automated question answering through reusing accumulated question–answer data. The framework, named QSem, defines three types of question words – question-target words, user-oriented words and irrelevant words, along with semantic patterns, for representing a question. The question word types are semantically labelled by a pre-defined ontology to enrich the semantic representation of questions. The semantic patterns through equivalent pattern linking enhance normal structure matching aiming at improving question matching performance. We trained QSem on 400 randomly selected questions with semantic patterns and obtained optimized parameters. After that, 5000 questions from our system were tested and the precision of question matching was between 0.71 and 0.93 with respect to various generators, indicating the stability of the approach. We further compared our approach with Cosine similarity, WordNet-based semantic similarity and IBM translation model on a standard TREC dataset containing 5536 questions. The results presented that our approach achieved best performance with mean reciprocal rank increased by 7.2% and accuracy increased by 7.5% on average, demonstrating the effectiveness of the approach.

Explaining human performance in psycholinguistic tasks with models of semantic similarity based on prediction and counting: A review and empirical validation

Abstract Recent developments in distributional semantics (Mikolov, Chen, Corrado, & Dean, 2013; Mikolov, Sutskever, Chen, Corrado, & Dean, 2013) include a new class of prediction-based models that are trained on a text corpus and that measure semantic similarity between words. We discuss the relevance of these models for psycholinguistic theories and compare them to more traditional distributional semantic models. We compare the models’ performances on a large dataset of semantic priming (Hutchison et al., 2013) and on a number of other tasks involving semantic processing and conclude that the prediction-based models usually offer a better fit to behavioral data. Theoretically, we argue that these models bridge the gap between traditional approaches to distributional semantics and psychologically plausible learning principles. As an aid to researchers, we release semantic vectors for English and Dutch for a range of models together with a convenient interface that can be used to extract a great number of semantic similarity measures.

Representational Similarity of Body Parts in Human Occipitotemporal Cortex.

Regions in human lateral and ventral occipitotemporal cortices (OTC) respond selectively to pictures of the human body and its parts. What are the organizational principles underlying body part responses in these regions? Here we used representational similarity analysis (RSA) of fMRI data to test multiple possible organizational principles: shape similarity, physical proximity, cortical homunculus proximity, and semantic similarity. Participants viewed pictures of whole persons, chairs, and eight body parts (hands, arms, legs, feet, chests, waists, upper faces, and lower faces). The similarity of multivoxel activity patterns for all body part pairs was established in whole person-selective OTC regions. The resulting neural similarity matrices were then compared with similarity matrices capturing the hypothesized organizational principles. Results showed that the semantic similarity model best captured the neural similarity of body parts in lateral and ventral OTC, which followed an organization in three clusters: (1) body parts used as action effectors (hands, feet, arms, and legs), (2) noneffector body parts (chests and waists), and (3) face parts (upper and lower faces). Whole-brain RSA revealed, in addition to OTC, regions in parietal and frontal cortex in which neural similarity was related to semantic similarity. In contrast, neural similarity in occipital cortex was best predicted by shape similarity models. We suggest that the semantic organization of body parts in high-level visual cortex relates to the different functions associated with the three body part clusters, reflecting the unique processing and connectivity demands associated with the different types of information (e.g., action, social) different body parts (e.g., limbs, faces) convey. Significance statement: While the organization of body part representations in motor and somatosensory cortices has been well characterized, the principles underlying body part representations in visual cortex have not yet been explored. In the present fMRI study we used multivoxel pattern analysis and representational similarity analysis to characterize the organization of body maps in human occipitotemporal cortex (OTC). Results indicate that visual and shape dimensions do not fully account for the organization of body part representations in OTC. Instead, the representational structure of body maps in OTC appears strongly related to functional-semantic properties of body parts. We suggest that this organization reflects the unique processing and connectivity demands associated with the different types of information different body parts convey.

Can the humped animal's knee conceal its name? Commentary on: "The roles of shared vs. distinctive conceptual features in lexical access".

Can the humped animal's knee conceal its name? Commentary on: "The roles of shared vs. distinctive conceptual features in lexical access".