Linguistics and Intellectual Technologies: Papers from the Annual Conference "Dialogue"

Introduction

This summary discusses three papers presented at the Annual Conference "Dialogue" on Linguistics and Intellectual Technologies. The conference took place in Moscow from May 27 to 30. The papers cover various topics related to linguistics and technology.

Paper 4: Semantic clustering of Russian web search results: possibilities and problems

This paper, authored by Kutuzov A., explores the concept of semantic clustering of Russian web search results. The author discusses the potential and challenges of using semantic clustering techniques in the context of Russian language search queries. The paper was presented at the Russian Summer School in Information Retrieval, held from August 18 to 22. The paper is published in the conference proceedings by Springer.

Paper 5: Adapting word2vec to named entity recognition

Sienčnik S.K. presents a paper on adapting word2vec, a popular word embedding model, to named entity recognition. The paper discusses the process of modifying word2vec to improve its performance in identifying named entities in text. The paper was presented at the 20th Nordic conference of computational linguistics, held from May 11 to 13 in Vilnius. The paper is published in the conference proceedings by Linköping University Electronic Press.

Paper 6: Vec2graph: A Python Library for Visualizing Word Embeddings as Graphs

Katricheva N., Yaskevich A., Lisitsina A., Zhordaniya T., Kutuzov A., and Kuzmenko E. present a paper on Vec2graph, a Python library for visualizing word embeddings as graphs. The authors discuss the features and capabilities of Vec2graph and its potential applications in analyzing word embeddings. The paper is published in the Communications in Computer and Information Science series by Springer. It is part of the proceedings of the AIST 2019 conference, which focused on the analysis of images, social networks, and texts.

Overall, these papers contribute to the field of linguistics and intellectual technologies by exploring various aspects of language processing, information retrieval, and visualization of word embeddings.

Date of Publication: May 30, 2019

Contact Information:

Keywords: Linguistics, Intellectual Technologies, Annual Conference, Dialogue, Semantic Clustering, Russian Web Search Results, Word2Vec, Named Entity Recognition, Vec2graph, Python Library, Word Embeddings, Graph Visualization, Language Processing, Information Retrieval, AIST 2019 Conference

Embeddings as Graphs

Introduction

The use of embeddings has become increasingly popular in various fields, including natural language processing and sentiment analysis. Embeddings are vector representations of words or entities that capture their semantic meaning. Traditionally, embeddings have been represented as vectors in a high-dimensional space, where the distance between vectors reflects the similarity between words. However, recent research has explored the idea of representing embeddings as graphs.

Embeddings as Graphs

The concept of embeddings as graphs involves representing embeddings as nodes in a graph, where the edges between nodes represent the relationships between embeddings. This approach allows for a more flexible and expressive representation of semantic relationships between words or entities. By representing embeddings as graphs, it becomes possible to capture not only the similarity between words, but also their hierarchical relationships, contextual associations, and other semantic connections.

Applications of Embeddings as Graphs

The use of embeddings as graphs has shown promise in various applications. One notable application is sentiment analysis, where the goal is to determine the sentiment or emotion expressed in a piece of text. By representing embeddings as graphs, it becomes possible to capture the complex relationships between words and their associated sentiments. This can lead to more accurate sentiment analysis models.

Another application is in the development of domain ontologies. A domain ontology is a formal representation of the concepts and relationships within a specific domain. By representing embeddings as graphs, it becomes possible to automatically generate or refine domain ontologies based on the semantic relationships between words or entities.

Related Work

The idea of representing embeddings as graphs builds upon previous research in the field. One notable work is the paper by Maas et al. (2011), where they proposed the use of word vectors for sentiment analysis. Their approach involved representing words as vectors in a high-dimensional space, where the distance between vectors reflected the similarity between words.

Another related work is the paper by Palagin et al. (2011), where they presented a technique for designing a domain ontology. Their approach involved using computer means, networks, and systems to automatically generate a domain ontology based on the semantic relationships between words or entities.

Conclusion

The concept of embeddings as graphs offers a new perspective on representing semantic relationships between words or entities. By representing embeddings as nodes in a graph, it becomes possible to capture not only the similarity between words, but also their hierarchical relationships, contextual associations, and other semantic connections. This approach has shown promise in various applications, including sentiment analysis and the development of domain ontologies. Further research is needed to explore the full potential of embeddings as graphs and to develop more advanced models and algorithms based on this concept.

Published on May 30, 2019

Publication source

See the PDF from which this article has been generated:

PDF source url: https://ceur-ws.org/Vol-2866/ceur_342-352palagin34.pdf