Mejora en la atención al cliente usando datos de Twitter y técnicas de aprendizaje automático
Loading...
Date
Authors
relationships.isAdvisorOf
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Torcuato Di Tella
Abstract
El comercio electrónico ha experimentado una aceleración sin precedentes en el último
tiempo. Por su parte, las empresas han intentado adaptarse para satisfacer la creciente
demanda de consumo brindando, en su mayoría, atención al cliente bajo una modalidad de
"autoservicio". De hecho, varias de ellas han utilizado redes sociales como medio de soporte
en pos de construir una imagen de disponibilidad e inmediatez. No obstante, los agentes
acaban recibiendo los problemas más complejos, dejando en evidencia la falta de
capacitación y herramientas provistos para la tarea. A lo largo de este trabajo, se diseña un
tablero de negocio para dar visibilidad del estado de situación respecto a la atención
brindada así como, también, para poder responder ante distintos casos de atención al cliente
de forma data-driven en un marco de recursos limitados. Se hace foco, particularmente, en
Mercado Libre ya que cuenta con un volumen considerable de transacciones y brinda
soporte vía Twitter. Recolectando datos de esta fuente, se explora y modela la probabilidad
de requerir atención al cliente, la satisfacción del usuario, las palabras clave del texto, los
tópicos y la ubicación del tweet. Para ello, se aplican técnicas de aprendizaje supervisado
como no supervisado, se emplean ensambles y se utilizan expresiones regulares.
Online commerce has recently experienced an unprecedented acceleration. For their part, companies have tried to adapt to meet the growing consumer demand by providing, for the most part, customer service under a "self-service" modality. In fact, several of them have used social networks as a means of support in order to build an image of availability and immediacy. However, the agents end up receiving the most complex problems, revealing the lack of training and tools provided for the task. Throughout this work, a business dashboard is designed to give visibility of the status of the service provided, as well as to be able to respond to different customer cases in a data-driven manner within a limited resources framework. The focus is particularly on Mercado Libre since it has a considerable volume of transactions and provides support via Twitter. Collecting data from this source, it is explored and modeled the probability of requiring customer service, user satisfaction, the keywords of the text, the topics and the location of the tweet. For this, supervised and unsupervised learning techniques are applied, ensembles are designed and regular expressions are used.
Online commerce has recently experienced an unprecedented acceleration. For their part, companies have tried to adapt to meet the growing consumer demand by providing, for the most part, customer service under a "self-service" modality. In fact, several of them have used social networks as a means of support in order to build an image of availability and immediacy. However, the agents end up receiving the most complex problems, revealing the lack of training and tools provided for the task. Throughout this work, a business dashboard is designed to give visibility of the status of the service provided, as well as to be able to respond to different customer cases in a data-driven manner within a limited resources framework. The focus is particularly on Mercado Libre since it has a considerable volume of transactions and provides support via Twitter. Collecting data from this source, it is explored and modeled the probability of requiring customer service, user satisfaction, the keywords of the text, the topics and the location of the tweet. For this, supervised and unsupervised learning techniques are applied, ensembles are designed and regular expressions are used.
Description
Keywords
Data Analysis, Predicción tecnológica, Comercio electrónico, Consumer behavior, Comportamiento del Consumidor, Análisis de datos