Document Type

Theses, Masters

Rights

Available under a Creative Commons Attribution Non-Commercial Share Alike 4.0 International Licence

Disciplines

1.2 COMPUTER AND INFORMATION SCIENCE

Publication Details

Dissertation submitted for the degree of Masters by research in Computing, Technological University Dublin, 2022.

Abstract

The work presented in this thesis was born from the desire to map documents with similar semantic concepts between them. We decided to address this problem as a named entity recognition task, where we have identified key concepts in the texts we use, and we have categorized them. So, we can apply named entity recognition techniques and automatically recognize these key concepts inside other documents. However, we propose the use of a classification method based on the recognition of named entities or key phrases, where the method can detect similarities between key concepts of the texts to be analyzed, and through the use of Poincaré embeddings, the model can associate the existing relationship between these concepts. Thanks to the Poincaré Embeddings’ ability to capture relationships between words, we were able to implement this feature in our classifier. Consequently for each word in a text we check if there are words close to it that are also close to the words that make up the key phrases that we use as Gold Standard. Therefore when detecting potential close words that make up a named entity, the classifier then applies a series of characteristics to classify it.

The methodology used performed better than when we only considered the POS structure of the named entities and their n-grams. However, determining the POS structure and the n-grams were important to improve the recognition of named entities in our research. By improving time to recognize similar key phrases between documents, some common tasks in large companies can have a notorious benefit. An important example is the evaluation of resumes, to determine the best professional for a specific position. This task is characterized by consuming a lot of time to find the best profiles for a position, but our contribution in this research work considerably reduces that time, finding the best profiles for a job. Here the experiments are shown considering job descriptions and real resumes, and the methodology used to determine the representation of each of these documents through their key phrases is explained.

DOI

https://doi.org/10.21427/f1s5-2c94


Share

COinS