Desarrollo de motores de búsqueda de productos en plataforma ecommerce utilizando procesamiento de lenguaje natural

Abstract

Since the beginnings of artificial intelligence until today there have been great technological advances, what was said to be science fiction is now reality. with the lapse of time the human being has become very ambitious so it has been acquiring various knowledge and enhancing them to create systems that help facilitate their daily lives, as is the case of robots that are created in order to do actions that humans perform, the development of smart phones, virtual assistants, among others which facilitate communication. During several generations humanity has had to adapt to different disasters that threaten its existence, one of those inconveniences is the pandemic caused by the covid19 that forced the population to remain in quarantine, in this way the activities were reduced, one of the sectors that had a remarkable growth is the online commerce because users began to visit these sites to make their purchases. E-commerce platforms focus more on design, forgetting an important point which are basic functions such as the product search engine, which is of great importance because users enter the online store, search for products, often obtaining ambiguous results or a lot of irrelevant information. Therefore, to fulfill the purpose of this project, we have proposed the development of a prototype of a search engine for e-commerce platforms using natural language processing. Natural language processing (NLP) is very important within artificial intelligence and NLP is also transcendental within computational linguistics, this allows to convert natural language into a mechanism that computers can understand. To conclude with the purpose of the project, a prototype architecture was developed, this will allow the processing of data phase by phase. Organizing the input data to apply natural language processing, and as a result obtaining an answer. The coding was done by using different tools: Java, Netbeans, JSP, Web Services and the Apache Lucene library; in addition, different methods and classes were developed, which combined with the methods of the Lucene library, perform a syntactic search. The evaluation stage was performed using classification metrics: precision and completeness or Recall. In the first evaluation a product query was performed, where the results were of the maximum value 100%, whereby the precision and sensitivity score were close to 1, this indicates that the response values are relevant with perfect accuracy. In the second test, a query with several words was performed where the results obtained in terms of precision were 33%, sensitivity 50%, which indicates that the values in the search are repeated in several products, which shows that the precision is not exact and some answers are not relevant. In the third evaluation, an incomplete data was entered to know the level of proximity of the search engine, which showed that the results returned in the query are correct in some cases.