Using Interpretive Semantics Techniques to Enhance Ontology Learning

Escobar-Vega, Luis M.

View/Open

PhDEngScITESOTesis.pdf (1.342Mb)

Date

2021-06

Author

Escobar-Vega, Luis M.

Metadata

Show full item record

URI

https://hdl.handle.net/11117/7461

Description

As intelligent virtual assistant scales to the mass market, traditional validation techniques for question answering systems become inappropriate to get full functional coverage of the system. Natural language and conversational dialog inherent complexities introduce design challenges to guarantee process, talk, and understanding performance. Besides, there is and an increasing number of training language models in question-answering systems. A significant portion of them corresponds to the statistic-based language model. Improvements in datasets, natural language processing techniques, and processing speed have allowed better data rates to scale beyond 90% of the Score. Some effects of the lack of interpretation can create multiple understanding integrity problems in solving a question. This problem is aggravated when the model faces a new and different context from that used in the training process. Challenges for meaning comprehension are continuously increasing. Therefore, information retrieval processes extract key elements of the language that can be critical for making more useful question-answering systems. Using appropriate information retrieval techniques to extract critical elements that can be used to create new knowledge structures is a significant challenge. The combination of information retrieval and ontology learning can be a very consuming validation task. Typical practices in question-answering systems construction are statistic-based. Consequently, they require massive datasets to train their models, making the information retrieval process too lengthy and prohibitive when the model faces new contexts. In this doctoral dissertation, the combination of interpretive semantics, semantic similarity, and ontology learning methods with suitable statistical functions is proposed to improve the efficiency of extracting semantic elements from a text. The proposed methods are implemented in a software tool, and its performance is evaluated on real question-answering platforms such as virtual assistants. The results show both the efficiency of the proposed methods and significant improvements when compared to state-of-the-art practices.

Collections

Documentos - ITESO