A method and device for identifying opinions in securities research reports

Abstract

The present invention provides a method and device for identifying opinions in securities research reports, and relates to the field of artificial intelligence technology. The method comprises: extracting feature words and opinion words in securities research reports, obtaining opinion words and opinion tendencies corresponding to each feature word; obtaining feature scores for each feature word according to the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words; obtaining opinion identification results for the securities research report according to the feature scores of each feature word and a research report opinion identification model; wherein the research report opinion identification model is obtained based on research report prediction training data and corresponding label training. The device is used to execute the above method. The method and device for identifying opinions in securities research reports provided by the embodiments of the present invention improve the accuracy of identifying opinions in securities research reports.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for identifying viewpoints in securities research reports. Background Art

[0002] At present, the number of securities research reports is increasing day by day, covering different topics such as macro research, industry research, company research, fund research, etc. By analyzing securities research reports, we can timely understand the situation of the industry, policies, investment feasibility, etc.

[0003] Faced with a huge number of securities research reports, investors have limited time and energy, and the number of securities research reports that can be read and analyzed is limited. In order to help investors quickly understand the views of securities research reports, in the prior art, securities research reports are divided into sentences, and the polarity of each sentence is determined by sentence-level viewpoint classification. The polarity is positive, negative or neutral, and the viewpoint of the securities research report is identified by the polarity of each sentence. Sentence-level polarity classification usually uses supervised learning algorithms. Applying supervised learning methods to subjective sentence-level classification requires manual labeling, which has low modeling efficiency. The analysis of securities research reports through sentence-level classification has low accuracy. Summary of the invention

[0004] In view of the problems in the prior art, an embodiment of the present invention provides a method and device for identifying opinions in securities research reports, which can at least partially solve the problems in the prior art.

[0005] In one aspect, the present invention provides a method for identifying viewpoints in a securities research report, comprising:

[0006] Extract the characteristic words and opinion words of the securities research report, and obtain the opinion words and the opinion tendency of the opinion words corresponding to each characteristic word;

[0007] According to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words, a feature score of each feature word is obtained;

[0008] According to the feature scores of each feature word and the research report opinion recognition model, the opinion recognition result of the securities research report is obtained; wherein, the research report opinion recognition model is obtained based on the research report prediction training data and the corresponding label training.

[0009] In another aspect, the present invention provides a device for identifying viewpoints in a securities research report, comprising:

[0010] An extraction module is used to extract the characteristic words and opinion words of the securities research report, and obtain the opinion words and the opinion tendency of the opinion words corresponding to each characteristic word;

[0011] An acquisition module, used for obtaining a feature score of each feature word according to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words;

[0012] The recognition module is used to obtain the opinion recognition result of the securities research report based on the feature scores of each feature word and the research report opinion recognition model; wherein the research report opinion recognition model is obtained based on the research report prediction training data and corresponding label training.

[0013] On the other hand, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the method for identifying opinions in securities research reports described in any of the above-mentioned embodiments.

[0014] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for identifying opinions in securities research reports described in any of the above-mentioned embodiments.

[0015] The method and device for identifying opinions in securities research reports provided in an embodiment of the present invention extract feature words and opinion words in securities research reports, obtain opinion words and opinion tendencies corresponding to each feature word, obtain feature scores for each feature word based on the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words, obtain opinion identification results for securities research reports based on the feature scores of each feature word and a research report opinion identification model, perform opinion identification based on opinion words and feature words, with finer granularity, and improve the accuracy of opinion identification in securities research reports. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0017] Figure 1 It is a flowchart of the method for identifying opinions in securities research reports provided by the first embodiment of the present invention.

[0018] Figure 2 It is a flowchart of the method for identifying opinions in securities research reports provided by the second embodiment of the present invention.

[0019] Figure 3 It is a flowchart of a method for identifying opinions in a securities research report provided in the third embodiment of the present invention.

[0020] Figure 4 It is a flowchart of the method for identifying opinions in securities research reports provided by the fourth embodiment of the present invention.

[0021] Figure 5 It is a flowchart of the method for identifying opinions in securities research reports provided by the fifth embodiment of the present invention.

[0022] Figure 6 It is a schematic diagram of the structure of the device for identifying opinions in securities research reports provided by the sixth embodiment of the present invention.

[0023] Figure 7 It is a schematic diagram of the structure of the device for identifying opinions in securities research reports provided by the seventh embodiment of the present invention.

[0024] Figure 8 It is a schematic diagram of the structure of the device for identifying opinions in securities research reports provided in the eighth embodiment of the present invention.

[0025] Fig. 9 It is a schematic diagram of the structure of the device for identifying opinions in securities research reports provided by the ninth embodiment of the present invention.

[0026] Fig.10 It is a schematic diagram of the structure of a device for identifying opinions in securities research reports provided in the tenth embodiment of the present invention.

[0027] Fig.11 It is a schematic diagram of the physical structure of an electronic device provided in the eleventh embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily.

[0029] Figure 1 FIG. 1 is a flow chart of a method for identifying viewpoints in a securities research report provided by the first embodiment of the present invention. Figure 1 As shown, the method for identifying viewpoints in a securities research report provided by an embodiment of the present invention includes:

[0030] S101, extracting characteristic words and opinion words from the securities research report, and obtaining the opinion words and the opinion tendency of the opinion words corresponding to each characteristic word;

[0031] Specifically, the server can obtain a securities research report, and then extract feature words and opinion words from the securities research report, establish a correspondence between the feature words and the opinion words, and determine the opinion tendency of the opinion words. Among them, opinion words are, for example, words such as daily limit, daily limit, big rise, big fall, and general rise. Feature words are objects described by opinion words. Opinion tendency refers to the emotional tendency expressed by opinion words, which can include positive, negative, and neutral. The executor of the opinion identification method for securities research reports provided in the embodiment of the present invention includes but is not limited to a server.

[0032] For example, in a securities research report including content such as "In terms of industry sectors, the new energy vehicle industry chain has soared across the board, and the electronics sector has performed well", the server can extract soaring and good as opinion words. The soaring describes the new energy vehicle industry, and the good describes the electronics sector. The server can extract the new energy vehicle industry and the electronics sector as feature words. The new energy vehicle industry corresponds to the soaring, and the electronics sector corresponds to the good. The opinion tendencies of the soaring and the good are both positive.

[0033] S102, obtaining a feature score for each feature word according to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words;

[0034] Specifically, the server will perform a feature score on each feature word according to the number of opinion words corresponding to each feature word in the securities research report and the opinion tendency of the opinion words corresponding to each feature word to obtain a feature score for each feature word.

[0035] For example, the server may count the number of opinion words with positive opinion tendencies among the opinion words corresponding to the feature words to obtain a valid number, and then calculate the ratio of the valid number to the number of opinion words corresponding to the feature words as the feature score of the feature words.

[0036] S103. Obtain the opinion recognition result of the securities research report based on the feature score of each feature word and the research report opinion recognition model; wherein the research report opinion recognition model is obtained based on the research report prediction training data and corresponding label training.

[0037] Specifically, the server inputs the feature scores of each feature word of the securities research report into the research report opinion recognition model, identifies the opinion of the securities research report, and obtains the opinion recognition result of the securities research report. The recognition result is positive or negative. If the recognition result is positive, it indicates that the securities research report is optimistic about the investment in securities and securities-related products and holds a recommendation attitude; if the recognition result is negative, it indicates that the securities research report is pessimistic about the investment in securities and securities-related products and holds a non-recommendation attitude. Among them, the research report opinion recognition model is obtained based on the research report prediction training data and the corresponding label training. The label is preset, and the label can include positive and negative.

[0038] The method for identifying opinions in securities research reports provided in an embodiment of the present invention extracts feature words and opinion words in securities research reports, obtains opinion words and opinion tendencies corresponding to each feature word, obtains feature scores for each feature word based on the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words, obtains opinion identification results for securities research reports based on the feature scores of each feature word and a research report opinion identification model, and performs opinion identification based on opinion words and feature words with finer granularity, thereby improving the accuracy of opinion identification in securities research reports.

[0039] Figure 2 FIG. 1 is a flow chart of a method for identifying viewpoints in a securities research report provided by a second embodiment of the present invention. Figure 2 As shown, based on the above embodiments, further, extracting the characteristic words and opinion words of the securities research report, and obtaining the opinion words and the opinion tendency of the opinion words corresponding to each characteristic word includes:

[0040] S201, segmenting and parsing the securities research report to obtain the part of speech and dependency relationship of each word in each sentence;

[0041] Specifically, the server can perform word segmentation and analysis on the securities research report, divide each sentence in the securities research report into individual words, and determine the part of speech of each word, as well as the dependency relationship between the words in each sentence. The part of speech includes nouns, verbs, adjectives, etc., and the dependency relationship includes the modification relationship between words.

[0042] For example, the server may segment and parse the securities research report using a natural language processing tool, wherein the natural language processing tool includes but is not limited to Standford CoreNLP.

[0043] S202, determining potential opinion words and potential feature words of each sentence according to the part of speech of each word of each sentence;

[0044] Specifically, since the feature words are nouns, the opinion words can be adjectives, adverbs and verbs, while prepositions, conjunctions, auxiliary words, interjections, onomatopoeias, quantifiers and pronouns will not be used as feature words and opinion words. The server will use the nouns in each sentence of the securities research report as potential feature words and the adjectives, adverbs and verbs in each sentence as potential opinion words according to the part of speech of each word.

[0045] S203. Determine each feature word and the corresponding opinion word and the opinion tendency of the opinion word according to each potential opinion word and potential feature word, the seed opinion vocabulary set and the dependency relationship between each word until no new opinion word appears; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word.

[0046] Specifically, the server traverses each potential opinion word, compares each potential opinion word with each seed opinion word in the seed opinion vocabulary set, and determines whether there is a seed opinion word identical to the potential opinion word in the seed opinion vocabulary set. If there is a seed opinion word identical to the potential opinion word, the potential opinion word is used as an opinion word, and the opinion tendency of the seed opinion word identical to the opinion word is used as the opinion tendency of the opinion word. Then, according to the dependency relationship of each word, it is determined whether the word corresponding to the opinion word is a potential feature word. If the word corresponding to the opinion word is a potential feature word, the potential feature word corresponding to the opinion word is used as a feature word.

[0047] After traversing all potential opinion words, the server will obtain each feature word (the first screening of feature words). Then, based on the dependency relationship between each feature word and each vocabulary, the potential opinion word corresponding to each feature word is obtained, and the potential opinion word corresponding to each feature word is used as the opinion word corresponding to each feature word. If one of the opinion words corresponding to the feature word is a seed opinion word, then the corresponding opinion tendency of the seed opinion word is obtained as the opinion tendency corresponding to the opinion word. If one of the opinion words corresponding to the feature word is not a seed opinion word, then the opinion tendency of the seed opinion word corresponding to the feature word is used as the opinion tendency corresponding to all opinion words corresponding to the feature word, that is, in the securities research report, the opinion tendencies corresponding to all opinion words corresponding to the feature word are consistent. The opinion words corresponding to each feature word are compared with each seed opinion word in the seed opinion vocabulary set, and the opinion words that are not in the seed opinion vocabulary set are found as new opinion words (the first screening of new opinion words), and the next feature word screening is performed. If no new opinion words are obtained, then this step ends.

[0048] When performing the second screening of feature words, the server traverses the new opinion words screened for the first time, and obtains the potential feature words corresponding to each opinion word in the first screening of new opinion words based on the dependency relationship between each opinion word and each word in the first screening of new opinion words, and then compares it with the first screening feature words to obtain potential feature words different from the first screening feature words as new feature words (second screening feature words). Then, based on each feature word in the second screening feature words and the dependency relationship between each word, the server obtains the potential opinion words corresponding to each feature word in the second screening feature words, and then compares it with the seed opinion words in the seed opinion vocabulary set and the first screening new opinion words to obtain new opinion words (second screening new opinion words), and perform the next screening of feature words. If no new opinion words are obtained for the second screening, this step ends.

[0049] By analogy, the server continuously iterates to obtain new opinion words until no new opinion words appear, and the server obtains each feature word and the corresponding opinion word and the opinion tendency of the opinion word. Among them, the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word. The number of seed opinion words included in the seed opinion vocabulary set is set according to actual needs, and the embodiment of the present invention does not limit it.

[0050] It should be noted that the opinion tendency of the seed opinion word is preset. When mining feature words through seed opinion vocabulary, it can be determined whether the opinion tendency of the feature word corresponding to the seed opinion word is positive, negative or neutral. Then the opinion tendency of the new opinion word mined by the feature word is determined according to the opinion tendency of the corresponding feature word. If a feature word in a securities research report corresponds to different seed opinion words, and the opinion tendencies of different seed opinion words are inconsistent, it can be considered that the feature word has no clear opinion tendency and can be considered invalid.

[0051] On the basis of the above embodiments, further, the method for identifying the opinions of securities research reports provided by the embodiments of the present invention also includes:

[0052] The seed opinion vocabulary set is updated according to the newly appeared opinion words and the corresponding opinion tendencies.

[0053] Specifically, after the server identifies opinions on the securities research report, if it obtains newly emerged opinion words and opinion tendencies corresponding to the newly emerged opinion words, it can add the newly emerged opinion words and opinion tendencies corresponding to the newly emerged opinion words to the seed opinion vocabulary set to expand the seed opinion vocabulary set.

[0054] It is understandable that in order to ensure the accuracy of the newly emerged opinion words and the corresponding opinion tendencies, before updating the seed opinion vocabulary set, the opinion words and the corresponding opinion tendencies to be added to the seed opinion vocabulary set may be manually checked.

[0055] Figure 3 : is a flow chart of a method for identifying viewpoints in a securities research report provided by a third embodiment of the present invention. Figure 3 As shown, on the basis of the above embodiments, further, the step of obtaining the research report opinion recognition model based on the research report prediction training data and the corresponding label training includes:

[0056] S301, extracting characteristic words and opinion words from each sample securities research report, and obtaining opinion words and opinion tendencies of the opinion words corresponding to the characteristic words of each sample securities research report;

[0057] Specifically, the server may obtain a preset number of sample securities research reports, and then extract the characteristic words and opinion words of each sample securities research report from each sample securities research report, establish a corresponding relationship between the characteristic words and opinion words of each sample securities research report, and determine the opinion tendency of the opinion words of each sample securities research report. The preset number is set according to actual needs and is not limited in the embodiment of the present invention.

[0058] S302, obtaining a feature score for each feature word of each sample securities research report based on the number of opinion words corresponding to each feature word of each sample securities research report and the opinion tendency of the opinion words, and using the feature score of each feature word of each sample securities research report as a sample data of the research report prediction training data and obtaining the opinion of each sample securities research report as a label of the corresponding sample data;

[0059] Specifically, the server will perform a feature score on each feature word of the sample securities research report according to the number of opinion words corresponding to each feature word of the sample securities research report and the opinion tendency of the opinion words corresponding to each feature word, and obtain the feature score of each feature word of the sample securities research report. The server uses the feature score of each feature word of each sample securities research report as a sample data of the research report prediction training data, and the server also obtains the opinion of each sample securities research report as the label of the corresponding sample data. Among them, the opinion of the sample securities research report can be obtained by experts after analyzing the sample securities research report, and the opinion of the sample securities research report includes positive and negative opinions.

[0060] S303. According to each sample data and corresponding labels of the research report prediction training data, as well as the initial model, the research report opinion recognition model is trained to obtain the research report opinion recognition model.

[0061] Specifically, the server can train the initial model according to each sample data and corresponding labels of the research report prediction training data to obtain the research report opinion recognition model. During training, the research report prediction training data can be divided into a training set and a verification set. The research report opinion recognition model to be verified is trained through the training set, and then the research report opinion recognition model to be verified is verified through the verification set. The research report opinion recognition model to be verified that passes the verification is used as the research report opinion recognition model. If the research report opinion recognition model to be verified fails to pass the verification, the model can be retrained by adding training data, modifying the parameters of the initial model, etc. Among them, the initial model can be a model established by the extreme gradient boosting (XGBoost) machine learning algorithm, or it can be a model established by a neural network algorithm. It can be selected according to actual needs, and the embodiment of the present invention does not limit it.

[0062] Figure 4 : is a flow chart of a method for identifying viewpoints in a securities research report provided by a fourth embodiment of the present invention. Figure 4 As shown, based on the above embodiments, further, extracting the characteristic words and opinion words of each sample securities research report, and obtaining the opinion words and the opinion tendency of the opinion words corresponding to the characteristic words of each sample securities research report includes:

[0063] S401, segmenting and parsing the sample securities research report to obtain the part of speech and dependency relationship of each word in each sentence of the sample securities research report;

[0064] Specifically, the server can perform word segmentation and analysis on the sample securities research report, divide each sentence in the sample securities research report into individual words, and determine the part of speech of each word, as well as the dependency relationship between the individual words in each sentence. The specific implementation process of this step is similar to step S201, and will not be repeated here.

[0065] S402, determining potential opinion words and potential feature words for each sentence of the sample securities research report according to the part of speech of each word in each sentence of the sample securities research report;

[0066] Specifically, the server will use the nouns in each sentence of the sample securities research report as potential feature words and the adjectives, adverbs and verbs in each sentence as potential opinion words according to the part of speech of each word. The specific implementation process of this step is similar to step S202 and will not be repeated here.

[0067] S403. Determine each feature word and the corresponding opinion word and the opinion tendency of the opinion word according to each potential opinion word and potential feature word of the sample securities research report, the seed opinion vocabulary set and the dependency relationship between each word, until no new opinion words appear; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word.

[0068] Specifically, the specific implementation process of this step is similar to step S203 and will not be repeated here.

[0069] On the basis of the above embodiments, further, the initial model adopts an extreme gradient boosting model. The extreme gradient boosting model is a model established based on the XGBoost machine learning algorithm. The demand for training samples of the extreme gradient boosting model is less than that of other machine learning algorithm models, and the efficiency of model training is higher.

[0070] Figure 5 : is a flow chart of a method for identifying viewpoints in a securities research report provided by a fifth embodiment of the present invention. Figure 5 As shown, based on the above embodiments, further, obtaining the feature score of each feature word according to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion word includes:

[0071] S501, counting the number of opinion words with positive opinion tendency among the opinion words corresponding to the characteristic words to obtain a valid number;

[0072] Specifically, the server may count the number of opinion words with positive opinion tendencies among the opinion words corresponding to the feature words according to the opinion tendencies corresponding to each opinion word among the opinion words corresponding to the feature words, as the effective number.

[0073] S502: Calculate the ratio of the effective number to the number of opinion words corresponding to the feature word as the feature score of the feature word.

[0074] Specifically, the server can count the number of opinion words included in the opinion words corresponding to the feature word as the number of opinion words corresponding to the feature word, and then calculate the ratio of the effective number to the number of opinion words corresponding to the feature word, and use the above ratio as the feature score of the feature word.

[0075] The opinion identification method for securities research reports provided in an embodiment of the present invention performs feature mining on securities research reports, identifies corresponding feature words and opinion words, and calculates the feature score corresponding to each feature word, and then identifies the opinion of the securities research report through the feature score of each feature word and the research report opinion identification model. Compared with the existing technology, it utilizes more information from the research report, has finer granularity, and has more accurate prediction results, which can help investors make correct decisions in the financial market and protect investors' returns to the greatest extent.

[0076] Figure 6 : is a schematic diagram of the structure of a device for identifying viewpoints in a securities research report provided by a sixth embodiment of the present invention. Figure 6 As shown, the viewpoint identification device of the securities research report provided by the embodiment of the present invention includes an extraction module 601, an acquisition module 602 and an identification module 603, wherein:

[0077] The extraction module 601 is used to extract the feature words and opinion words of the securities research report, and obtain the opinion words and the opinion tendencies of the opinion words corresponding to each feature word; the acquisition module 602 is used to obtain the feature score of each feature word based on the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words; the identification module 603 is used to obtain the opinion recognition results of the securities research report based on the feature scores of each feature word and the research report opinion recognition model; wherein the research report opinion recognition model is obtained based on the research report prediction training data and the corresponding label training.

[0078] Specifically, the extraction module 601 can obtain a securities research report, and then extract feature words and opinion words from the securities research report, establish a correspondence between the feature words and the opinion words, and determine the opinion tendency of the opinion words. Among them, opinion words are, for example, words such as daily limit, daily limit, big rise, big fall, and general rise. Feature words are objects described by opinion words. Opinion tendency refers to the emotional tendency expressed by opinion words, which can include positive, negative, and neutral.

[0079] The acquisition module 602 will perform a feature score on each feature word according to the number of opinion words corresponding to each feature word in the securities research report and the opinion tendency of the opinion words corresponding to each feature word, so as to obtain the feature score of each feature word.

[0080] The identification module 603 inputs the feature scores of each feature word of the securities research report into the research report opinion identification model, identifies the opinion of the securities research report, and obtains the opinion identification result of the securities research report. The identification result is positive or negative. If the identification result is positive, it indicates that the securities research report is optimistic about the investment in securities and securities-related products and holds a recommendation attitude; if the identification result is negative, it indicates that the securities research report is pessimistic about the investment in securities and securities-related products and holds a non-recommendation attitude. Among them, the research report opinion identification model is obtained based on the research report prediction training data and the corresponding label training. The label is preset, and the label can include positive and negative.

[0081] The device for identifying opinions in securities research reports provided in an embodiment of the present invention extracts feature words and opinion words in securities research reports, obtains opinion words and opinion tendencies corresponding to each feature word, obtains feature scores for each feature word based on the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words, obtains opinion identification results for securities research reports based on the feature scores of each feature word and a research report opinion identification model, and performs opinion identification based on opinion words and feature words with finer granularity, thereby improving the accuracy of opinion identification in securities research reports.

[0082] Figure 7 : is a schematic diagram of the structure of a device for identifying viewpoints in a securities research report provided by a seventh embodiment of the present invention. Figure 7 As shown, based on the above embodiments, further, the extraction module 601 includes a parsing unit 6011, a first determining unit 6012 and a second determining unit 6013, wherein:

[0083] The parsing unit 6011 is used to segment and parse the securities research report to obtain the part of speech and dependency relationship of each word in each sentence; the first determination unit 6012 is used to determine the potential opinion words and potential feature words of each sentence according to the part of speech of each word in each sentence; the second determination unit 6013 is used to determine each feature word and the corresponding opinion word and the opinion tendency of the opinion word according to each potential opinion word and potential feature word, the seed opinion vocabulary set and the dependency relationship of each word, until no new opinion word appears; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word.

[0084] On the basis of the above embodiments, further, the method for identifying the opinions of securities research reports provided by the embodiments of the present invention further includes an updating module:

[0085] The updating module is used to update the seed opinion vocabulary set according to the newly appeared opinion words and the corresponding opinion tendencies.

[0086] Figure 8 : is a schematic diagram of the structure of a device for identifying viewpoints in a securities research report provided by an eighth embodiment of the present invention. Figure 8 As shown, based on the above embodiments, further, the viewpoint identification device of the securities research report provided by the embodiment of the present invention further includes a sample extraction module 604, a training data acquisition module 605 and a training module 606, wherein:

[0087] The sample extraction module 604 is used to extract the feature words and opinion words of the sample securities research reports, and obtain the opinion words and the opinion tendencies of the opinion words corresponding to the feature words of each sample securities research report; the training data acquisition module 605 is used to obtain the feature score of each feature word of each sample securities research report based on the number of opinion words corresponding to each feature word of each sample securities research report and the opinion tendencies of the opinion words, and use the feature score of each feature word of each sample securities research report as a sample data of the research report prediction training data and obtain the opinion of each sample securities research report as the label of the corresponding sample data; the training module 606 is used to train a research report opinion recognition model based on the sample data and corresponding labels of the research report prediction training data, as well as the initial model.

[0088] Fig. 9 : is a schematic diagram of the structure of a viewpoint identification device for a securities research report provided by a ninth embodiment of the present invention. Fig. 9 As shown, based on the above embodiments, further, the sample extraction module 604 includes a sample parsing unit 6041, a third determination unit 6042 and a fourth determination unit 6043, wherein:

[0089] The sample parsing unit 6041 is used to segment and parse the sample securities research report to obtain the part of speech and dependency relationship of each word in each sentence of the sample securities research report; the third determination unit 6042 is used to determine the potential opinion words and potential feature words of each sentence according to the part of speech of each word in each sentence of the sample securities research report; the fourth determination unit 6043 traverses the potential opinion words and potential feature words of the sample securities research report according to the seed opinion vocabulary set and the dependency relationship of each word in the sample securities research report, and determines each feature word of the sample securities research report and the corresponding opinion words and the opinion tendency of the opinion words until no new feature words or opinion words appear; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word.

[0090] On the basis of the above embodiments, further, the initial model adopts an extreme gradient boosting model.

[0091] Fig.10: is a schematic diagram of the structure of a device for identifying viewpoints in a securities research report provided by a tenth embodiment of the present invention. Fig.10 As shown, based on the above embodiments, further, the obtaining module 602 includes a statistical unit 6021 and a calculation unit 6022, wherein:

[0092] The statistical unit 6021 is used to count the number of opinion words with positive opinion tendencies among the opinion words corresponding to the feature words to obtain the effective number; the calculation unit 6022 is used to calculate the ratio of the effective number to the number of opinion words corresponding to the feature words as the feature score of the feature words.

[0093] The embodiments of the apparatus provided in the embodiments of the present invention can be specifically used to execute the processing flow of the above-mentioned method embodiments, and its functions are not described in detail here, and reference can be made to the detailed description of the above-mentioned method embodiments.

[0094] Fig.11 is a schematic diagram of the physical structure of an electronic device provided by the eleventh embodiment of the present invention, such as Fig.11 As shown, the electronic device may include: a processor 1101, a communication interface 1102, a memory 1103 and a communication bus 1104, wherein the processor 1101, the communication interface 1102 and the memory 1103 communicate with each other through the communication bus 1104. The processor 1101 may call the logic instructions in the memory 1103 to execute the following method: extract the feature words and opinion words of the securities research report, obtain the opinion words and the opinion tendency of the opinion words corresponding to each feature word; obtain the feature score of each feature word according to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words; obtain the opinion recognition result of the securities research report according to the feature scores of each feature word and the research report opinion recognition model; wherein the research report opinion recognition model is obtained based on the research report prediction training data and the corresponding label training.

[0095] In addition, the logic instructions in the above-mentioned memory 1103 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0096] The present embodiment discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: extracting feature words and opinion words from a securities research report, obtaining the opinion words and the opinion tendencies of the opinion words corresponding to each feature word; obtaining a feature score for each feature word based on the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words; obtaining the opinion recognition result of the securities research report based on the feature scores of each feature word and a research report opinion recognition model; wherein the research report opinion recognition model is obtained based on research report prediction training data and corresponding label training.

[0097] The present embodiment provides a computer-readable storage medium, which stores a computer program, and the computer program enables the computer to execute the methods provided by the above-mentioned method embodiments, for example, including: extracting feature words and opinion words from the securities research report, and obtaining the opinion words and the opinion tendencies of the opinion words corresponding to each feature word; obtaining the feature score of each feature word based on the number of opinion words corresponding to each feature word and the opinion tendencies of the opinion words; obtaining the opinion recognition result of the securities research report based on the feature scores of each feature word and the research report opinion recognition model; wherein the research report opinion recognition model is obtained based on the research report prediction training data and the corresponding label training.

[0098] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0100] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0101] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0102] In the description of this specification, the description with reference to the terms "one embodiment", "a specific embodiment", "some embodiments", "for example", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0103] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying opinions in securities research reports. It is characterized in that include: Extract the characteristic words and opinion words of the securities research report, and obtain the opinion words and opinion tendencies corresponding to each characteristic word, including: Segment and parse the securities research report to obtain the part of speech and dependency relationship of each word in each sentence; Determine the potential opinion words and potential feature words of each sentence according to the parts of speech of each word in each sentence; According to the dependency relationship between each potential opinion word and potential feature word, the seed opinion vocabulary set and each word, each feature word and the corresponding opinion word and the opinion tendency of the opinion word are determined until no new opinion word appears; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word; specifically comprising: traversing each potential opinion word, comparing each potential opinion word with each seed opinion word in the seed opinion vocabulary set, judging whether there is a seed opinion word identical to the potential opinion word in the seed opinion vocabulary set, if there is a seed opinion word identical to the potential opinion word, then the potential opinion word is used as an opinion word, and the opinion tendency of the seed opinion word identical to the opinion word is used as the opinion tendency of the opinion word; then according to the dependency relationship between each word, determining whether the word corresponding to the opinion word is a potential feature word, if the word corresponding to the opinion word is a potential feature word, then the potential feature word corresponding to the opinion word is used as a feature word; After traversing all potential opinion words, the first screening feature word set is obtained; Based on the dependency relationship between each feature word and each word, a potential opinion word corresponding to each feature word is obtained, and the potential opinion word corresponding to each feature word is used as the opinion word corresponding to each feature word. If one of the opinion words corresponding to the feature word is a seed opinion word, then the opinion tendency corresponding to the seed opinion word is obtained as the opinion tendency corresponding to the opinion word. If one of the opinion words corresponding to the feature word is not a seed opinion word, then the opinion tendency of the seed opinion word corresponding to the feature word is used as the opinion tendency corresponding to all the opinion words corresponding to the feature word. Compare the opinion words corresponding to each feature word with each seed opinion word in the seed opinion vocabulary set, find out the opinion words that are not in the seed opinion vocabulary set as the first screening of new opinion word sets, and perform the next feature word screening; if no new opinion words are obtained, then this step ends; When performing the second feature word screening, the first screened new opinion word set is traversed, and according to the dependency relationship between each opinion word and each word in the first screened new opinion word set, the potential feature word corresponding to each opinion word in the first screened new opinion word set is obtained, and then compared with the first screened feature word set, and the potential feature words different from the first screened feature word set are obtained as the second screened feature word set; Based on each feature word in the second screening feature word set and the dependency relationship of each word, obtain the potential opinion word corresponding to each feature word in the second screening feature word set, and then compare with the seed opinion words in the seed opinion vocabulary set and the first screening new opinion word set to obtain the second screening new opinion word set, and perform the next feature word screening; if the second screening new opinion word set is not obtained, then this step ends; In this way, a new opinion word set is obtained by continuous iteration until no new opinion word set appears, and each feature word and the corresponding opinion word and opinion tendency of the opinion word are obtained; According to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words, a feature score of each feature word is obtained; According to the feature scores of each feature word and the research report opinion recognition model, the opinion recognition result of the securities research report is obtained; wherein, the research report opinion recognition model is obtained based on the research report prediction training data and the corresponding label training.

2. The method according to claim 1, It is characterized in that Also includes: The seed opinion vocabulary set is updated according to the newly appeared opinion words and the corresponding opinion tendencies.

3. The method according to claim 1, It is characterized in that The steps of obtaining a research report opinion recognition model based on research report prediction training data and corresponding label training include: Extract the characteristic words and opinion words of each sample securities research report, and obtain the opinion words and the opinion tendency of the opinion words corresponding to the characteristic words of each sample securities research report; According to the number of opinion words corresponding to each feature word of each sample securities research report and the opinion tendency of the opinion words, a feature score of each feature word of each sample securities research report is obtained, and the feature score of each feature word of each sample securities research report is used as a sample data of the research report prediction training data, and the opinion of each sample securities research report is obtained as a label of the corresponding sample data; According to each sample data and corresponding labels of the research report prediction training data, as well as the initial model, the research report opinion recognition model is trained to obtain the research report opinion recognition model.

4. The method according to claim 3, It is characterized in that The extracting of the characteristic words and opinion words of each sample securities research report, and obtaining the opinion words and the opinion tendency of the opinion words corresponding to the characteristic words of each sample securities research report include: Segmenting and parsing the sample securities research report to obtain the part of speech and dependency relationship of each word in each sentence of the sample securities research report; Determine potential opinion words and potential feature words for each sentence of the sample securities research report according to the part of speech of each word in each sentence of the sample securities research report; According to the potential opinion words and potential feature words of the sample securities research report, the seed opinion vocabulary set and the dependency relationship between each word, each feature word and the corresponding opinion word and the opinion tendency of the opinion word are determined until no new opinion words appear; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word.

5. The method according to claim 3, It is characterized in that The initial model adopts the extreme gradient boosting model.

6. The method according to any one of claims 1 to 5, It is characterized in that The step of obtaining a feature score for each feature word according to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words includes: Counting the number of opinion words with positive opinion tendency among the opinion words corresponding to the feature words to obtain a valid number; The ratio of the effective number to the number of opinion words corresponding to the feature word is calculated as the feature score of the feature word.

7. A device for identifying opinions in securities research reports, It is characterized in that include: An extraction module is used to extract the characteristic words and opinion words of the securities research report, and obtain the opinion words and the opinion tendency of the opinion words corresponding to each characteristic word; The extraction module includes a parsing unit, a first determining unit and a second determining unit, wherein: The parsing unit is used to segment and parse the securities research report to obtain the part of speech and dependency relationship of each word in each sentence; The first determination unit is used to determine the potential opinion words and potential feature words of each sentence according to the part of speech of each word of each sentence; The second determination unit is used to determine each feature word and the corresponding opinion word and the opinion tendency of the opinion word according to the dependency relationship between each potential opinion word and potential feature word, the seed opinion vocabulary set and each word, until no new opinion word appears; wherein the seed opinion vocabulary set is preset, including multiple seed opinion words and the opinion tendency corresponding to each seed opinion word; specifically comprising: traversing each potential opinion word, comparing each potential opinion word with each seed opinion word in the seed opinion vocabulary set, judging whether there is a seed opinion word identical to the potential opinion word in the seed opinion vocabulary set, if there is a seed opinion word identical to the potential opinion word, then the potential opinion word is used as an opinion word, and the opinion tendency of the seed opinion word identical to the opinion word is used as the opinion tendency of the opinion word; then, according to the dependency relationship between each word, determining whether the word corresponding to the opinion word is a potential feature word, if the word corresponding to the opinion word is a potential feature word, then the potential feature word corresponding to the opinion word is used as a feature word; After traversing all potential opinion words, the first screening feature word set is obtained; Based on the dependency relationship between each feature word and each word, a potential opinion word corresponding to each feature word is obtained, and the potential opinion word corresponding to each feature word is used as the opinion word corresponding to each feature word. If one of the opinion words corresponding to the feature word is a seed opinion word, then the opinion tendency corresponding to the seed opinion word is obtained as the opinion tendency corresponding to the opinion word. If one of the opinion words corresponding to the feature word is not a seed opinion word, then the opinion tendency of the seed opinion word corresponding to the feature word is used as the opinion tendency corresponding to all the opinion words corresponding to the feature word. Compare the opinion words corresponding to each feature word with each seed opinion word in the seed opinion vocabulary set, find out the opinion words that are not in the seed opinion vocabulary set as the first screening of new opinion word sets, and perform the next feature word screening; if no new opinion words are obtained, then this step ends; When performing the second feature word screening, the first screened new opinion word set is traversed, and according to the dependency relationship between each opinion word and each word in the first screened new opinion word set, the potential feature word corresponding to each opinion word in the first screened new opinion word set is obtained, and then compared with the first screened feature word set, and the potential feature words different from the first screened feature word set are obtained as the second screened feature word set; Based on each feature word in the second screening feature word set and the dependency relationship of each word, obtain the potential opinion word corresponding to each feature word in the second screening feature word set, and then compare with the seed opinion words in the seed opinion vocabulary set and the first screening new opinion word set to obtain the second screening new opinion word set, and perform the next feature word screening; if the second screening new opinion word set is not obtained, then this step ends; In this way, a new opinion word set is obtained by continuous iteration until no new opinion word set appears, and each feature word and the corresponding opinion word and opinion tendency of the opinion word are obtained; An acquisition module, used for obtaining a feature score of each feature word according to the number of opinion words corresponding to each feature word and the opinion tendency of the opinion words; The recognition module is used to obtain the opinion recognition result of the securities research report based on the feature scores of each feature word and the research report opinion recognition model; wherein the research report opinion recognition model is obtained based on the research report prediction training data and corresponding label training.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images