Automatic query expansion for vehicle repair documents through user behavior

The process of Information Retrieval (IR) by query driven search engines have become an essential part of the customer experience in any data related digital product. The accuracy and completeness of the search results is a matter of great interest and a crucial key performance indicator. An importa...

Descripción completa

Detalles Bibliográficos
Autor: Ghiringhelli, Juan Carlos
Tipo de recurso: tesis de maestría
Fecha de publicación:2020
País:España
Institución:Universitat Oberta de Catalunya (UOC)
Repositorio:O2, repositorio institucional de la UOC
OAI Identifier:oai:openaccess.uoc.edu:10609/121326
Acceso en línea:http://hdl.handle.net/10609/121326
Access Level:acceso abierto
Palabra clave:nlp
embeddings
query expansion
synonyms
pln
incorporaciones
expansión de consultas
sinónimos
incorporacions
expansió de consultes
sinònims
Artificial intelligence -- TFM
Intel·ligència artificial -- TFM
Inteligencia artificial -- TFM
Descripción
Sumario:The process of Information Retrieval (IR) by query driven search engines have become an essential part of the customer experience in any data related digital product. The accuracy and completeness of the search results is a matter of great interest and a crucial key performance indicator. An important enhancer for search engines is query expansion Query Expansion (QE), where equivalent search queries Equivalent Search Query (ESQ) are added to the original request to increase recall. ESQs can be discovered using the same tools as synonym discovery given certain considerations, taking advantage of the fact that synonym discovery is a well developed field of Natural Language Processing (NLP) with many available techniques. The motivation for this project is to use the tools available in NLP Machine Learning (ML) to automatically detect ESQs. For this a large sample of logs describing search query customer behavior was used. This data set was obtained from a live enterprise product that publishes repair documents for automobiles. Graph embeddings through an implementation method called node2Vec and vector cosine similarity is the chosen discovery method for the ESQs. The conclusion of the experiment is that while usable search expansion queries are discovered, extra human intervention or further automatic selection is necessary to filter the valuable cases from the large number of found cases, even working within a strict similarity threshold.