Automatic extraction method for abstracts based on public company announcements

Abstract

The invention relates to an automatic extraction method for abstracts based on public company announcements. The automatic extraction method for abstracts comprises the following steps: S1, obtaining public company announcement files from securities exchanges to form an announcement file database; S2, utilizing a word2vec model to obtain word vectors from a text corpus; S3, calculating out the similarity between sentences to construct a sentence graph model; S4, calculating out the weight of the sentences; S5, adjusting a sentence weight matrix according to sentence positions; S6, choosing sentences which are maximum in weight and free of redundancy to form an abstract. The automatic extraction method for abstracts based on the public company announcements can provide accurate abstract files with higher readability for financial market investors, help the investors to understand in a shorter time and well make investment judgments and also provide important indexes for quantitative fund companies.

Description

technical field [0001] The invention relates to the field of data extraction, in particular to an automatic extraction method based on an abstract of a listed company announcement. Background technique [0002] As of mid-June 2016, there were a total of 2,832 stocks in the Shanghai and Shenzhen stock markets, and hundreds to thousands of announcements were issued every day. With the rapid development of the Internet, the cost of editing is getting lower and lower, the dissemination of information is faster and faster, and the number of daily announcements is also increasing rapidly. At present, the announcements of listed companies are generally lengthy and the terminology is professional. However, most investors in China are retail investors and do not have enough time to read the announcements carefully. Moreover, it is difficult for ordinary investors to quickly identify the important content and make reasonable judgments. Therefore, it is very important and valuable to ...

AI Technical Summary

Problems solved by technology

However, the method based on graph sorting has defects in calculating the similarity between sentences, ignoring the unique characteristics of listed company announcements

The title of a listed company’s announcement document often contains a lot of key information, so a sentence with a high degree of similarity to the title of the announcement is more likely to become an abstract, and this sentence will have a greater impact on the surrounding sentences. In addition, the listed company’s announcement It often contains a lot of key terms (restructuring, allotment, repurchase, additional issuance, net profit, increase or decrease over the same period, risk, etc.), and the sentences containing these key terms are more likely to become abstracts. Of course, company announcements are often more standardized. Therefore, the position of the sentence in the paragraph also contains a lot of information, so the method of only using the similarity between sentences does not take into account these many factors, which affects the accuracy of the node sentence weight and the accuracy of the formed summary. sex

Images