Semantic Similarity Measures Research Articles

A reproducible survey on word embeddings and ontology-based methods for word similarity: Linear combinations outperform the state of the art

Human similarity and relatedness judgements between concepts underlie most of cognitive capabilities, such as categorisation, memory, decision-making and reasoning. For this reason, the proposal of methods for the estimation of the degree of similarity and relatedness between words and concepts has been a very active line of research in the fields of artificial intelligence, information retrieval and natural language processing among others. Main approaches proposed in the literature can be categorised in two large families as follows: (1) Ontology-based semantic similarity Measures (OM) and (2) distributional measures whose most recent and successful methods are based on Word Embedding (WE) models. However, the lack of a deep analysis of both families of methods slows down the advance of this line of research and its applications. This work introduces the largest, reproducible and detailed experimental survey of OM measures and WE models reported in the literature which is based on the evaluation of both families of methods on a same software platform, with the aim of elucidating what is the state of the problem. We show that WE models which combine distributional and ontology-based information get the best results, and in addition, we show for the first time that a simple average of two best performing WE models with other ontology-based measures or WE models is able to improve the state of the art by a large margin. In addition, we provide a very detailed reproducibility protocol together with a collection of software tools and datasets as supplementary material to allow the exact replication of our results.

Combining and learning word embedding with WordNet for semantic relatedness and similarity measurement

In this research, we propose 3 different approaches to measure the semantic relatedness between 2 words: (i) boost the performance of GloVe word embedding model via removing or transforming abnormal dimensions; (ii) linearly combine the information extracted from WordNet and word embeddings; and (iii) utilize word embedding and 12 linguistic information extracted from WordNet as features for Support Vector Regression. We conducted our experiments on 8 benchmark data sets, and computed Spearman correlations between the outputs of our methods and the ground truth. We report our results together with 3 state‐of‐the‐art approaches. The experimental results show that our method can outperform state‐of‐the‐art approaches in all the selected English benchmark data sets.

A word-embedding-based approach for accurate identification of corresponding activities

Smart process repositories have been developed to effectively manage large collections of business process models. Several key features of such repositories require accurate matching of corresponding activities between a pair of process models. However, the existing matching techniques have not yet achieved the desired accuracy, which impedes the effectiveness of process repositories. To that end, this paper proposes a word-embedding-based approach that significantly improves the accuracy of matching. For a comprehensive evaluation of the proposed approach, we have performed experiments using three state-of-the-art word embeddings, two syntactic measures, six semantic measures, and four datasets. The results show that the use of word embeddings outperforms all the syntactic as well as the semantic similarity measures. Moreover, the use of fastText-based embeddings in our proposed technique achieves the highest F1 score, compared to both Word2vec- and GloVe-based embeddings.

A novel method for merging academic social network ontologies using formal concept analysis and hybrid semantic similarity measure

PurposeThe purpose of this paper is to merge the ontologies that remove the redundancy and improve the storage efficiency. The count of ontologies developed in the past few eras is noticeably very high. With the availability of these ontologies, the needed information can be smoothly attained, but the presence of comparably varied ontologies nurtures the dispute of rework and merging of data. The assessment of the existing ontologies exposes the existence of the superfluous information; hence, ontology merging is the only solution. The existing ontology merging methods focus only on highly relevant classes and instances, whereas somewhat relevant classes and instances have been simply dropped. Those somewhat relevant classes and instances may also be useful or relevant to the given domain. In this paper, we propose a new method called hybrid semantic similarity measure (HSSM)-based ontology merging using formal concept analysis (FCA) and semantic similarity measure.Design/methodology/approachThe HSSM categorizes the relevancy into three classes, namely highly relevant, moderate relevant and least relevant classes and instances. To achieve high efficiency in merging, HSSM performs both FCA part and the semantic similarity part.FindingsThe experimental results proved that the HSSM produced better results compared with existing algorithms in terms of similarity distance and time. An inconsistency check can also be done for the dissimilar classes and instances within an ontology. The output ontology will have set of highly relevant and moderate classes and instances as well as few least relevant classes and instances that will eventually lead to exhaustive ontology for the particular domain.Practical implicationsIn this paper, a HSSM method is proposed and used to merge the academic social network ontologies; this is observed to be an extremely powerful methodology compared with other former studies. This HSSM approach can be applied for various domain ontologies and it may deliver a novel vision to the researchers.Originality/valueThe HSSM is not applied for merging the ontologies in any former studies up to the knowledge of authors.

Effective spatio-temporal semantic trajectory generation for similar pattern group identification

The daily trajectories of individual movements convey a concise overview of their behaviors, with different social roles having different trajectory patterns. Therefore, we can identify users or groups based on the similar of their trajectory patterns. However, most existing trajectory analysis focuses only on the spatial and temporal analyses of the raw trajectory data and misses essential semantic information concerning behaviors. In this paper, we propose a new trajectory semantics calculation method to identify groups with similar behaviors. We first propose a fast and efficient two-phase method for identifying stay regions within daily trajectories and enriching the stay regions with semantic labels based on points of interest to generate semantic trajectories. Furthermore, we design a semantic similarity measure model using geographic and semantic similarity factors to measure the similarity between semantic trajectories. We also propose a pruning strategy using time entropy to decrease the number of complex calculations and comparisons to improve performance. The results of our extensive experiments on the real trajectory dataset of the Geolife project show that our proposed method is both effective and efficient.

A New Hypred Improved Method for Measuring Concept Semantic Similarity in WordNet

Computing semantic similarity between concepts is an important issue in natural language processing, artificial intelligence, information retrieval and knowledge management. The measure of computing concept similarity is a fundament of semantic computation. In this paper, we analyze typical semantic similarity measures and note Wu and Palmer’s measure which does not distinguish the similarities between nodes from a node to different nodes of the same level. Then, we synthesize the advantages of measure of path-based and IC-based, and propose a new hybrid method for measuring semantic similarity. By testing on a fragment of WordNet hierarchical tree, the results demonstrate the proposed method accurately distinguishes the similarities between nodes from a node to different nodes of the same level and overcome the shortcoming of the Wu and Palmer’s measure

Crime base: Towards building a knowledge base for crime entities and their relationships from online news papers

In the current era of internet, information related to crime is scattered across many sources namely news media, social networks, blogs, and video repositories, etc. Crime reports published in online newspapers are often considered as reliable compared to crowdsourced data like social media and contain crime information not only in the form of unstructured text but also in the form of images. Given the volume and availability of crime-related information present in online newspapers, gathering and integrating crime entities from multiple modalities and representing them as a knowledge base in machine-readable form will be useful for any law enforcement agencies to analyze and prevent criminal activities. Extant research works to generate the crime knowledge base, does not address extraction of all non-redundant entities from text and image data present in multiple newspapers. Hence, this work proposes Crime Base, an entity relationship based system to extract and integrate crime related text and image data from online newspapers with a focus towards reducing duplicity and loss of information in the knowledge base. The proposed system uses a rule-based approach to extract the entities from text and image captions. The entities extracted from text data are correlated using contextual as-well-as semantic similarity measures and image entities are correlated using low-level and high-level image features. The proposed system also presents an integrated view of these entities and their relations in the form of a knowledge base using OWL. The system is tested for a collection of crime related articles from popular Indian online newspapers.

Leveraging synonymy and polysemy to improve semantic similarity assessments based on intrinsic information content

Semantic similarity measures based on the estimation of the information content (IC) of concepts are currently regarded as the state of the art. Calculating the IC in an intrinsic (i.e., ontology-based) way is particularly convenient due to its accuracy and lack of dependency on annotated corpora. Intrinsic IC calculation models estimate concept probabilities from the taxonomic knowledge (i.e., number of hyponyms and/or hypernyms of the concepts) modelled in an ontology. In this paper, we aim to improve the intrinsic calculation of the IC by leveraging not only the hyponyms and hypernyms of concepts, but also the explicit evidences of synonymy and polysemy that ontologies such as WordNet also model. Specifically, we propose a more accurate intrinsic estimation of the concepts’ probabilities in which the IC calculation relies. We evaluate the accuracy of our proposal through a set of comprehensive experiments in which our IC calculation model is tested on a variety of IC-based similarity measures and benchmarks. Experimental results show that our proposal obtains consistently good accuracies, which vary less across measures and benchmarks than the most prominent intrinsic IC calculation models available in the literature.

Toward semantic similarity measure between concepts in an ontology

<p>A similarity measure is one classical problem in Description Logic which aims at identifying the similarity between concepts in an ontology. Finding a hierarchy distance among concepts in an ontology is one popular technique. However, one major drawback of such a technique is that it usually ignores a concept definition analysis. This work introduces a new method for similarity measure. The proposed system semantically analyzes structures of two concept descriptions and then computes the similarity score based on the number of shared features. The efficiency of the proposed algorithm is measured by means of the satisfaction of desirable properties and intensive experiments on the Snomed ct ontology.</p>

Semantic similarity aggregators for very short textual expressions: a case study on landmarks and points of interest

Semantic similarity measurement aims to automatically compute the degree of similarity between two textual expressions that use different representations for naming the same concepts. However, very short textual expressions cannot always follow the syntax of a written language and, in general, do not provide enough information to support proper analysis. This means that in some fields, such as the processing of landmarks and points of interest, results are not entirely satisfactory. In order to overcome this situation, we explore the idea of aggregating existing methods by means of two novel aggregation operators aiming to model an appropriate interaction between the similarity measures. As a result, we have been able to improve the results of existing techniques when solving the GeReSiD and the SDTS, two of the most popular benchmark datasets for dealing with geographical information.

Story co-segmentation of Chinese broadcast news using weakly-supervised semantic similarity

This paper presents lexical story co-segmentation, a new approach to automatically extracting stories on the same topic from multiple Chinese broadcast news documents. Unlike topic tracking and detection, our approach needs not the guidance of well-trained topic models and can consistently segment the common stories from input documents. Following the MRF scheme, we construct a Gibbs energy function that feasibly balances the intra-doc and inter-doc lexical semantic dependencies and solve story co-segmentation as a binary labeling problem at sentence level. Due to the significance of measuring lexical semantic similarity in story co-segmentation, we propose a weakly-supervised correlated affinity graph (WSCAG) model to effectively derive the latent semantic similarities between Chinese words from the target corpus. Based on this, we are able to extend the classical cosine similarity by mapping the observed words distribution into the latent semantic space, which leads to a generalized lexical cosine similarity measurement. Extensive experiments on benchmark dataset validate the effectiveness of our story co-segmentation approach. Besides, we specifically demonstrate the superior performance of the proposed WSCAG semantic similarity measure over other state-of-the-art semantic measures in story co-segmentation.

Multi-domain semantic similarity in biomedical research

BackgroundGiven the increasing amount of biomedical resources that are being annotated with concepts from more than one ontology and covering multiple domains of knowledge, it is important to devise mechanisms to compare these resources that take into account the various domains of annotation. For example, metabolic pathways are annotated with their enzymes and their metabolites, and thus similarity measures should compare them with respect to both of those domains simultaneously.ResultsIn this paper, we propose two approaches to lift existing single-ontology semantic similarity measures into multi-domain measures. The aggregative approach compares domains independently and averages the various similarity values into a final score. The integrative approach integrates all the relevant ontologies into a single one, calculating similarity in the resulting multi-domain ontology using the single-ontology measure.ConclusionsWe evaluated the two approaches in a multidisciplinary epidemiology dataset by evaluating the capacity of the similarity measures to predict new annotations based on the existing ones. The results show a promising increase in performance of the multi-domain measures over the single-ontology ones in the vast majority of the cases. These results show that multi-domain measures outperform single-domain ones, and should be considered by the community as a starting point to study more efficient multi-domain semantic similarity measures.

English semantic feature production norms: An extended database of 4436 concepts.

A limiting factor in understanding memory and language is often the availability of large numbers of stimuli to use and explore in experimental studies. In this study, we expand on three previous databases of concepts to over 4000 words including nouns, verbs, adjectives, and other parts of speech. Participants in the study were asked to provide lists of features for each concept presented (a semantic feature production task), which were combined with previous research in this area. These feature lists for each concept were then coded into their root word form and affixes (i.e., cat and s for cats) to explore the impact of word form on semantic similarity measures, which are often calculated by comparing concept feature lists (feature overlap). All concept features, coding, and calculated similarity information is provided in a searchable database for easy access and utilization for future researchers when designing experiments that use word stimuli. The final database of word pairs was combined with the Semantic Priming Project to examine the relation of semantic similarity statistics on semantic priming in tandem with other psycholinguistic variables.

Automatic design of semantic similarity controllers based on fuzzy logics

Recent advances in machine learning have been able to make improvements over the state-of-the-art regarding semantic similarity measurement techniques. In fact, we have all seen how classical techniques have given way to promising neural techniques. Nonetheless, these new techniques have a weak point: they are hardly interpretable. For this reason, we have oriented our research towards the design of strategies being able to be accurate enough but without sacrificing their interpretability. As a result, we have obtained a strategy for the automatic design of semantic similarity controllers based on fuzzy logics, which are automatically identified using genetic algorithms (GAs). After an exhaustive evaluation using a number of well-known benchmark datasets, we can conclude that our strategy fulfills both expectations: it is able of achieving reasonably good results, and at the same time, it can offer high degrees of interpretability.

In silico markers: an evolutionary and statistical approach to select informative genes of human breast cancer subtypes.

Recent advancement in bioinformatics offers the ability to identify informative genes from high dimensional gene expression data. Selection of informative genes from these large datasets has emerged as an issue of major concern among researchers. Gene functionality and regulatory mechanisms can be understood through the analysis of these gene expression data. Here, we present a computational method to identify informative genes for breast cancer subtypes such as Basal, human epidermal growth factor receptor 2 (Her2), luminal A (LumA), and luminal B (LumB). The proposed In Silico Markers method is a wrapper feature selection method based on Least Absolute Shrinkage and Selection Operator (LASSO), Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and Support Vector Machine (SVM) as a classifier. Moreover, the composite measure consisting of relevance, redundancy, and rank score of frequently appeared genes are used to select informative genes. The informative genes are validated by statistical and biologically relevant criteria. For a comparative evaluation of the proposed approach, biological similarity score designed on semantic similarity measure of GO terms are investigated. Further, the proposed technique is evaluated with 7 existing gene selection techniques using two-class annotated breast cancer subtype datasets. The utilization of this method can bring about the discovery of informative genes. Furthermore, under multiple criteria decision-making set-up, informative genes selected by the In Silico Markers are found to be admirable than the compared methods selected genes.

A review on semantic similarity measures for ontology

A review on semantic similarity measures for ontology

Predicting disease-related phenotypes using an integrated phenotype similarity measurement based on HPO

BackgroundImproving efficiency of disease diagnosis based on phenotype ontology is a critical yet challenging research area. Recently, Human Phenotype Ontology (HPO)-based semantic similarity has been affectively and widely used to identify causative genes and diseases. However, current phenotype similarity measurements just consider the annotations and hierarchy structure of HPO, neglecting the definition description of phenotype terms.ResultsIn this paper, we propose a novel phenotype similarity measurement, termed as DisPheno, which adequately incorporates the definition of phenotype terms in addition to HPO structure and annotations to measure the similarity between phenotype terms. DisPheno also integrates phenotype term associations into phenotype-set similarity measurement using gene and disease annotations of phenotype terms.ConclusionsCompared with five existing state-of-the-art methods, DisPheno shows great performance in HPO-based phenotype semantic similarity measurement and improves the efficiency of disease identification, especially on noisy patients dataset.

GO functional similarity clustering depends on similarity measure, clustering method, and annotation completeness

BackgroundBiological knowledge, and therefore Gene Ontology annotation sets, for human genes is incomplete. Recent studies have reported that biases in available GO annotations result in biased estimates of functional similarities of genes, but it is still unclear what the effect of incompleteness itself may be, even in the absence of bias. Pairwise gene similarities are used in a number of contexts, including gene “functional similarity” clustering and the related problem of functional ontology structure inference, but it is not known how different similarity measures or clustering methods perform on this task, and how the clusters are affected by annotation completeness.ResultsWe developed representations of both “complete” and “incomplete” GO annotation datasets based on experimentally-supported annotations from the GO database—specifically designed to model the incompleteness of human gene annotations—and computed semantic similarities for each set using a variety of different published measures. We then assessed the clusters derived from these measures using two different clustering methods: hierarchical clustering, and the CliXO algorithm. We find the CliXO algorithm, combined with appropriate measures, performs better than hierarchical clustering in reconstructing GO both when the data are complete, and incomplete. Some measures, particularly those that create a pairwise gene similarity by averaging over all pairwise annotation similarities, had consistently poor performance, and a few measures, such as Lin best-matched average and Relevance maximum, were generally among the best performers for a broad range in annotation completeness and types of GO classes. Finally, we show that for semantic similarity-based clustering, the multicellular organism process branch of the GO biological process ontology is more challenging to represent than the cellular process branch.ConclusionsWe assessed the effects of annotation completeness on the distribution of pairwise gene semantic similarity scores, and subsequent effects on the clusters derived from these scores. Our results suggest combinations of semantic similarity measures, gene-level scoring methods and clustering method that perform best for functional gene clustering using annotation sets of varying completeness. Overall, our results underscore the importance of increasing the completeness of GO annotations to for supporting computational analyses of gene function.

Semantic textual similarity between sentences using bilingual word semantics

Semantic textual similarity between sentences is indispensable for many information retrieval tasks. Traditional lexical similarity measures cannot compute the similarity beyond a trivial level. Moreover, they only can capture the textual similarity, but not semantic. In this paper, we propose a method for semantic textual similarity that leverages bilingual word-level semantics to compute the semantic similarity between sentences. To capture word-level semantics, we employ distribute representation of words in two different languages. The similarity function based on the concept-to-concept relationship corresponding to the words is also utilized for the same purpose. Multiple new semantic similarity measures are introduced based on word-embedding models trained on two different corpora in two different languages. Apart from these, another new semantic similarity measure is also introduced using the word sense comparison. The similarity score between the sentences is then computed by applying a linear ranking approach to all proposed measures with their importance score estimated employing a supervised feature selection technique. We conducted experiments on the SemEval Semantic Textual Similarity (STS-2017) test collections. The experimental results demonstrated that our method is effective for measuring semantic textual similarity and outperforms some known related methods.

Semantic similarity measures for formal concept analysis using linked data and WordNet

Formal Concept Analysis (FCA) is a field of applied mathematics with its roots in order theory, in particular the theory of complete lattices. It is not only a method for data analysis and knowledge representation, but also a formal formulation for concept formation and learning. Over the past 20 years, FCA has been widely studied. In this paper, the current research progresses and the existing problems of similarity measures in FCA are analyzed. To address the drawbacks of the existing methods, we propose a kind of novel semantic similarity measure for FCA by using Linked Data and WordNet. We aim to develop a method that is fully automatic without requiring predefined domain ontologies and can be used independently of the domain in applications requiring semantic similarity measures in FCA. To realize the semantic similarity estimation for FCA, we firstly extend the similarity assessment methods for resources (or entities) in Linked Data into semantic cases by using WordNet. Furthermore, we propose two kinds of semantic similarity measures (i.e., context-free method and context-aware method) for FCA concepts and concept lattices, respectively. Compared with the existing similarity measure methods in FCA, the proposed approach uses concept of possibility theory to determine lower and upper bounds of similarity intervals. Finally, we evaluate the proposed similarity assessment approaches by applying them to real-worlds datasets.