This paper presents a method for measuring the semantic similarity between concepts in Knowledge Graphs (KGs) such as WordNet and DBpedia. Previous work on semantic similarity methods have focused on either the structure of the semantic network between concepts (e.g., path length and depth), or only on the Information Content (IC) of concepts. We propose a semantic similarity method, namely wpath, to combine these two approaches, using IC to weight the shortest path length between concepts. Conventional corpus-based IC is computed from the distributions of concepts over textual corpus, which is required to prepare a domain corpus containing annotated concepts and has high computational cost. As instances are already extracted from textual corpus and annotated by concepts in KGs, graph-based IC is proposed to compute IC based on the distributions of concepts over instances. Through experiments performed on well known word similarity datasets, we show that the wpath semantic similarity method has produced a statistically significant improvement over other semantic similarity methods. Moreover, in a real category classification evaluation, the wpath method has shown the best performance in terms of accuracy and F score.

/pdf/computing-semantic-similarity-of-concepts-in-knowledge-hxtgnjme9z.pdf

Computing Semantic Similarity of Concepts in Knowledge Graphs

Linked Open Data has been recognized as a valuable source for background information in many data mining and information retrieval tasks. However, most of the existing tools require features in propositional form, i.e., a vector of nominal or numerical features associated with an instance, while Linked Open Data sources are graphs by nature. In this paper, we present RDF2Vec, an approach that uses language modeling approaches for unsupervised feature extraction from sequences of words, and adapts them to RDF graphs.We generate sequences by leveraging local information from graph sub-structures, harvested by Weisfeiler-Lehman Subtree RDF Graph Kernels and graph walks, and learn latent numerical representations of entities in RDF graphs.We evaluate our approach on three different tasks: (i) standard machine learning tasks, (ii) entity and document modeling, and (iii) content-based recommender systems. The evaluation shows that the proposed entity embeddings outperform existing techniques, and that pre-computed feature vector representations of general knowledge graphs such as DBpedia and Wikidata can be easily reused for different tasks.

/pdf/rdf2vec-rdf-graph-embeddings-and-their-applications-ejuhzgjfmy.pdf

RDF2Vec: RDF graph embeddings and their applications

Millennium Institute for Foundational Research on Data (IMFD)
Comision Nacional de Investigacion Cientifica y Tecnologica (CONICYT), CONICYT FONDECYT: 1181896

Information extraction meets the semantic web: a survey

With the increasing popularity of large scale Knowledge Graph (KG)s, many applications such as semantic analysis, search and question answering need to link entity mentions in texts to entities in KGs. Because of the polysemy problem in natural language, entity disambiguation is thus a key problem in current research. Existing disambiguation methods have considered entity prominence, context similarity and entity-entity relatedness to discriminate ambiguous entities, which are mainly working on document or paragraph level texts containing rich contextual information, and based on lexical matching for computing context similarity. When meeting short texts containing limited contextual information, such as web queries, questions and tweets, those conventional disambiguation methods are not good at handling single entity mention and measuring context similarity. In order to enhance the performance of disambiguation methods based on context similarity with such short texts, we propose SCSNED method for disambiguation based on semantic similarity between contextual words and informative words of entities in KGs. Specially, we exploit the effectiveness of both knowledge-based and corpus-based semantic similarity methods for entity disambiguation with SCSNED. Moreover, we propose a Category2Vec embedding model based on joint learning of word and category embedding, in order to compute word-category similarity for entity disambiguation. We show the effectiveness of these proposed methods with illustrative examples, and evaluate their effectiveness in a comparative experiment for entity disambiguation in real world web queries, questions and tweets. The experimental results have identified the effectiveness of different semantic similarity methods, and demonstrated the improvement of semantic similarity methods in SCSNED and Category2Vec over the conventional context similarity baseline. We further compare the proposed approaches with the state of the art entity disambiguation systems and show the performances of the proposed approaches are among the best performing systems. In addition, one important feature of the proposed approaches using semantic similarity, is the potential application on any existing KGs since they mainly use common features of entity descriptions and categories. Another contribution of the paper is an updated survey on background of entity disambiguation in KGs and semantic similarity methods.

Exploiting semantic similarity for named entity disambiguation in knowledge graphs

Providing relevant personalized recommendations for new users is one of the major challenges in recommender systems. This problem, known as the user cold start has been approached from different perspectives. In particular, cross-domain recommendation methods exploit data from source domains to address the lack of user preferences in a target domain. Most of the cross-domain approaches proposed so far follow the paradigm of collaborative filtering, and avoid analyzing the contents of the items, which are usually highly heterogeneous in the cross-domain setting. Content-based filtering, however, has been successfully applied in domains where item content and metadata play a key role. Such domains are not limited to scenarios where items do have text contents (e.g., books, news articles, scientific papers, and web pages), and where text mining and information retrieval techniques are often used. Potential application domains include those where items have associated metadata, e.g., genres, directors and actors for movies, and music styles, composers and themes for songs. With the advent of the Semantic Web, and its reference implementation Linked Data, a plethora of structured, interlinked metadata is available on the Web. These metadata represent a potential source of information to be exploited by content-based and hybrid filtering approaches. Motivated by the use of Linked Data for recommendation purposes, in this paper we present and evaluate a number of matrix factorization models for cross-domain collaborative filtering that leverage metadata as a bridge between items liked by users in different domains. We show that in case the underlying knowledge graph connects items from different domains and then in situations that benefit from cross-domain information, our models can provide better recommendations to new users while keeping a good trade-off between recommendation accuracy and diversity.

Addressing the user cold start with cross-domain collaborative filtering: exploiting item metadata in matrix factorization

Semantic relatedness and disambiguation are fundamental problems for linking text documents to the Web of Data. There are many approaches dealing with both problems but most of them rely on word or concept distribution over Wikipedia. They are therefore not applicable to concepts that do not have a rich textual description. In this paper, we show that semantic relatedness can also be accurately computed by analysing only the graph structure of the knowledge base. In addition, we propose a joint approach to entity and word-sense disambiguation that makes use of graph-based relatedness. As opposed to the majority of state-of-the-art systems that target mainly named entities, we use our approach to disambiguate both entities and common nouns. In our experiments, we first validate our relatedness measure on multiple knowledge bases and ground truth datasets and show that it performs better than related state-of-the-art graph based measures. Afterwards, we evaluate the disambiguation algorithm and show that it also achieves superior disambiguation accuracy with respect to alternative state-of-the-art graph-based algorithms.

/pdf/path-based-semantic-relatedness-on-linked-data-and-its-use-bqjnpk1nrx.pdf

Path-Based Semantic Relatedness on Linked Data and Its Use to Word and Entity Disambiguation

Online Social Media provides real-time information about events and news in the physical world. A challenging problem is then to identify in a timely manner the few relevant bits of information in these massive and fast-paced streams. Most of the current topic clustering and event detection methods focus on user generated content, hence they are sensible to language, writing style and are usually expensive to compute. Instead, our approach focuses on mining the structure of the graph generated by the interactions between users. Our hypothesis is that bursts in user interest for particular topics and events are reflected by corresponding changes in the structure of the discussion dynamics. We show that our method is capable of effectively identifying event topics in Twitter ground truth data, while offering better overall performance than a purely content-based method based on LDA topic models.

Graph-Based Methods for Clustering Topics of Interest in Twitter

In this paper, we propose a novel approach for social media event finding in order to support fast access to information that users find relevant While there are many approaches related to this problem, they mainly focus on homogeneous data, such as either the text of the posts, or the network of users Our research focuses on combining multiple types of data from social media in a heterogeneous network We propose different graph-based models using users, posts, and concepts extracted from the post content to represent the social media network We analyse the resulted heterogeneous network, and use it in order to cluster posts by different topics and events Our preliminary results show improvement over the methods that typically use only one type of data

Event Analysis in Social Media Using Clustering of Heterogeneous Information Networks

Entity relatedness is a task that is required in many applications such as entity disambiguation and clustering. Although there are many works on entity relatedness, most of them do not consider temporal aspects and thus, it is unclear how entity relatedness changes over time. This paper attempts to address this gap by showing how entity relatedness develops over time with graph-based approaches using Wikipedia and DBpedia as well as how transient links and stable links, by which is meant links that do not persist over different times and links that persist over time respectively, affect the relatedness of the entities. We show that different versions of the knowledge base at different times give different accuracy on the KORE dataset, and show how using multiple versions of the knowledge base through time can increase the accuracy of entity relatedness over a version of the knowledge graph from a single time point.

/pdf/temporal-evolution-of-entity-relatedness-using-wikipedia-and-3tl457f2ec.pdf

Temporal Evolution of Entity Relatedness using Wikipedia and DBpedia.

Princeton WordNet is the most widely-used lexical resource in natural language processing and continues to provide a gold standard model of semantics. However, there are still significant quality issues with the resource and these affect the performance of all NLP systems built on this resource. One major issue is that many nodes are insufficiently defined and new links need to be added to increase performance in NLP. We combine the use of graph-based metrics with measures of ambiguity in order to predict which synsets are difficult for word sense disambiguation, a major NLP task, which is dependent on good lexical information. We show that this method allows use to find poorly defined nodes with a 89.9% precision, which would assist manual annotators to focus on improving the most in-need parts of the WordNet graph.

Narumol Prangnawarat

Papers

Path-Based Semantic Relatedness on Linked Data and Its Use to Word and Entity Disambiguation

Graph-Based Methods for Clustering Topics of Interest in Twitter

Event Analysis in Social Media Using Clustering of Heterogeneous Information Networks

Temporal Evolution of Entity Relatedness using Wikipedia and DBpedia.

Identifying Poorly-Defined Concepts in WordNet with Graph Metrics