Research On Automatic Generation Of A-Share Review Text Based On Trend Word Discovery

With the rapid development of our economy and the gradual improvement of the financial supervision system,the stock market continues to expand and gradually tends to be stable.In the stock market,the daily stock market commentary as a market summary has attracted much attention from institutions and investors.Despite the rapid development of technology today,the collection,processing and creation of most stock market data and texts still require human expertise in the corresponding field.Therefore,if the automated generation of stock market commentary text can be realized,it will greatly liberate the productivity of financial practitioners,improve their work efficiency,so that they have more time and energy to focus on more meaningful work.However,the realization of automation still faces many challenges and problems to overcome.First of all,the creation of stock market comment text is based on the price change of the stock market in a day,so the generation of stock market comment is a Data-to-text problem.The traditional data-text generation benchmark data sets are mostly concentrated in weather,sports and other fields,where most of the text can be directly supported by data.However,compared with these fields,the correlation between data and text in the financial field is more complex.Moreover,there are more professional words in the text,which is not easy to collect.Second,early data-to-text generation methods were based on templates,which required manual production and were too costly to maintain as the variety of stock market commentary texts increased.Finally,many statements in stock market commentary texts require numerical reasoning,but current generation models do not perform well in these areas.Therefore,in view of the above problems,this thesis conducted an in-depth study,taking A-share as an example,and did the following work:(1)Discover trends in stock market data series.After obtaining abundant A-share minute data and corresponding stock market comment text,we found that the trend of the series corresponded to the trend words in the text after in-depth research,and the most important part of the stock market comment text was the trend words.(2)Construct stock market sequence and trend word rule base.Trend words are screened out from the stock market commentary texts obtained.After comparing the sequence and trend words,it can be seen that there are obvious rules between them.The derivative of the sequence is selected as the feature,which corresponds to the trend words one by one,and all the trend words and features are constructed together to form the rule base.(3)Set up stock market sequence to trend term rule model.The minute data of A-share A day is regarded as a time series curve.After inputting the data,the model can quickly match the corresponding trend words according to the curve characteristics.Considering that the numerical reasoning effect of the generated model is not good,and the words needed for numerical reasoning in the generated text are limited,this thesis uses the method of rules to solve this problem,which can assist the numerical reasoning of the generated model.(4)Build a generation model of A-share data to stock market commentary text.In this model,data segments are screened,important sequences are selected as model inputs,and trend word sequences are added,which are encoded by the bidirectional short and long time memory network.Fusion vectors are obtained and input into the generator to generate text.The experimental results show that the quality of the text generated by the model is high,and the accuracy of the text generated by the model to describe the trend change of data is significantly improved after adding the trend word sequence.