Discovery of Financial Advertising Subjects Based on Pre-training Model

Abstract

This paper mainly puts forward a text-oriented method of discovery of financial advertisement subjects, which can recognize the actual publishing organization of an Internet financial advertisement, thus be applied to the research of Internet financial advertisements. Differently from other methods, it divides the financial advertisement subject discovery into two sub tasks: named entity recognition in financial domain, as the core of this method, and enterprise or organization name matching. This paper adopts pre-training models such as BERT and ALBERT in named entity recognition method in the financial field, and trains six models based on pre-training models. Then name matching method is employed to obtain the advertising subjects. Specifically, it applies the best BERT + BiLSTM + CRF model tested in the first step to recognize the financial named entities contained in each target data in combination with data preprocessing and post-processing, determines the advertising subjects through the name matching of enterprises and institutions, and visually displays the results of advertising subjects. The method proposed in this paper is tested in financial advertising data from two sources. The results show that it has generality in recognizing the advertising subjects of financial advertising in different distribution. In theory, it can be applied to most of financial advertising, and achieve a high recognition rate on the data with high degree of standardization.