金融机构海量投诉数据分析与应用——基于LDA-TPA模型文本挖掘

Analysis of Massive Complaint Data of Financial Institutions Based on LDA-TPA model text Mining

近年来伴随新媒体的不断发展,金融服务行业的投诉话题极易暴露在社会公众视角下,成为舆论关注与讨论的焦点。据2020年金融监管部门投诉公开数据显示,金融领域投诉量正在以两位数逐年增长,如何有效挖掘海量投诉文本价值,洞悉投诉关注的主题及发展趋势,成为金融管理部门亟需解决的问题。本文在LDA模型基础上提出TPA算法^(①),通过实证分析某股份制银行2018年1月至2020年12月投诉数据库,识别出账单服务、卡片使用、业务营销、贷款申请等20个常见投诉主题,发现"贷款申请、还款方式、额度提升、费用减免"等方面的投诉主题增长趋势较为明显,呈现出周期性上升特点,而"商城购物、商品发货、券票使用"等方面的投诉主题应成为下一轮监管部门关注的焦点,为投诉数据文本挖掘及针对性改善金融服务体验提供方法论。

In recent years,with the continuous development of new media,the complaint topic of financial service industry is easily exposed to the perspective of the public and becomes the focus of public attention and discussion.According to the public complaint data of the financial regulatory department in 2020,the number of complaints in the financial field is increasing by double digits year by year.How to effectively dig the value of the massive complaint text and understand the theme and development trend of the complaint has become an urgent problem to be solved by the financial management department.Based on the LDA model,this paper proposes TPA algorithm,and identifies 20 common complaint topics such as billing service,card use,business marketing and loan application through empirical analysis of complaint database of a joint-stock bank from January 2018 to December 2020.Found that the loan application,repayment method,quota promotion and cost reduction"theme growth trends evident in the aspects such as complaints,present a cyclical upswing characteristics,and“shopping mall,goods delivery,coupon ticket use”complaints of theme should become the focus of attention of the next round of regulatory authorities,for the complaint data text mining provide methodology and targeted to improve financial services experience.