A method and system for constructing a causal graph based on code words

Abstract

The present invention provides a method and system for constructing a causal graph based on code words, the method comprising: constructing a causal graph in the financial field, and fusing the causal graph in the financial field with a knowledge graph in the financial field; constructing a code word library and a conversion dictionary matching the code word library; obtaining the semantics of the live content and judging whether it includes a first code word, if so, inputting the semantics of the live content into the causal graph in the financial field, and inferring whether the phenomenon input by the video audience in the barrage occurs later; identifying the second code word contained in the barrage, and performing operations on the barrage according to the emotional tendency of the second code word. The present invention intelligently identifies and analyzes the barrage containing code words in the live broadcast in the financial field, can display the popular semantics of the code word barrage to the audience in real time, and automatically shields the code word barrage containing negative and sensitive content, thereby improving the viewing experience of users.

Description

Technical Field

[0001] The present invention relates to the field of computer application technology, and in particular to a method and system for constructing a causal graph based on code words. Background Art

[0002] With the improvement of people's living standards in recent years, people are more willing to learn some financial knowledge and conduct some investment activities, and watching live broadcasts is a very important way for people to obtain financial knowledge. In the live broadcast of the financial field, in addition to the host's explanation and presentation, the barrage sent by netizens is also an important source for people to obtain relevant knowledge. However, in order to better express and communicate, the barrage sent by netizens often includes many code words. These code words are constantly created by people inadvertently, are commonly accepted in the industry, and can express related sentences. Therefore, these code words are difficult for ordinary people to understand, which reduces their user experience and increases the difficulty of the live broadcast platform to detect and review the barrage.

[0003] Invention patent CN201711293661.8 A method and system for constructing a knowledge graph of facts and principles, extracting event entities with preset types from structured data to form a macro event layer, forming an event evolution chain between event entities in the macro event layer through aggregate evolution, and extracting the causal relationship between event entities based on the event evolution chain; extracting micro entities from the structured data to form a micro knowledge layer, and mapping the causal relationship to the micro knowledge layer according to the mapping relationship between event entities and micro entities; abstracting and summarizing the causal relationship between micro entities and micro entities in the micro knowledge layer to form a knowledge ontology layer of facts and principles, and the macro event layer, micro knowledge layer and knowledge ontology layer of facts and principles constitute a knowledge graph of facts and principles. However, it does not take into account the code language problem existing in professional fields, and proposes a solution from this perspective.

[0004] The current causal graph system used in live broadcasts cannot identify and translate the coded comments into their usual meanings, nor can it correctly judge the emotional tendency of the coded comments. It can only analyze the coded comments according to ordinary words. As a result, it is very likely to make a wrong judgment on the emotional color of the coded comments, and thus make incorrect operations on the coded comments. Incorrect operations will greatly reduce the user experience and even cause user disgust, resulting in user loss. Therefore, it is necessary to rebuild a causal graph system based on coded comments to intelligently analyze and process the coded comments that appear in live broadcasts in the financial field to improve the user experience. Summary of the invention

[0005] The present invention provides a method for constructing a coin-based event graph, which mainly includes the following steps: constructing a financial field event graph, and integrating the financial field event graph with a financial field knowledge graph; constructing a coin-based word library and a conversion dictionary matching the coin-based word library;

[0006] The semantics of the live content is obtained and it is determined whether the first codeword is included. If so, the semantics of the live content is input into the financial field event map, and it is inferred whether the phenomenon input by the video viewer in the barrage will occur later; the second codeword included in the barrage is identified, and according to the emotional tendency of the second codeword, the operation of the barrage is performed.

[0007] Further optionally, in the above method, the step of constructing a financial field affairs graph and integrating it with a financial field knowledge graph includes:

[0008] Obtain massive amounts of text in the financial field on the Internet, and mine the logic of the massive amounts of text in the financial field; structure the logic of the matter through explicit patterns, integrate the logic of the matter from different sources, and abstract it layer by layer to obtain logic of the matter at different levels;

[0009] The constructed logic of events is linked by connecting the knowledge graph end to end to obtain the logic graph of the financial field based on events;

[0010] Entity recognition and entity linking technology are used to link the entities of events in the financial field event graph to the corresponding financial field knowledge graph, thereby realizing the fusion of the financial field event graph and the financial field knowledge graph.

[0011] Further optionally, in the method as described above, the logic of affairs includes: cause and effect affairs, conditional affairs, reversal affairs, sequential affairs, superior and subordinate affairs, composition affairs, and concurrent affairs; it also includes generalizing the emotional polarity of the event action itself to form an abstract cause and effect pattern chain.

[0012] Further optionally, in the method as described above, the step of constructing a codeword library and a conversion dictionary matching the codeword library includes:

[0013] Manually input words including codeword meanings to build a codeword word library;

[0014] Matching the first letters of the coded words in the coded word library with the words in the common dictionary, if the coded words in the coded word library match the words in the common dictionary successfully, determining the popular semantics of the coded words in the coded word library, if the match fails, matching the semantics of the coded words with the words in the candidate dictionary;

[0015] A conversion dictionary of popular semantics matched by the code words in the code word library is constructed.

[0016] Further optionally, in the method described above, the obtaining of the semantics of the live content and determining whether it includes codewords, if so, inputting the semantics of the live content into the financial field event graph, and inferring whether the phenomenon of video viewers inputting in the barrage occurs subsequently, including:

[0017] Through speech recognition based on pattern matching, the semantics of the content spoken by the anchor during the live broadcast is converted into the semantics of phrases or sentences;

[0018] Matching the semantics of the phrase or sentence with the codeword library, determining whether the semantics of the phrase or sentence contains a first similar codeword, and if so, inputting the first similar codeword into the financial field event graph to obtain events related to the first similar codeword;

[0019] Calculating a first semantic similarity of the context of the first similar codeword, and determining that the first similar codeword is a first codeword when the first semantic similarity is higher than a first preset threshold;

[0020] Input the semantics of the live broadcast content into the event graph to infer whether the phenomenon of video viewers inputting in the barrage will occur later.

[0021] Further optionally, in the method described above, the preset threshold of the first semantic similarity is 50%.

[0022] Further optionally, in the method described above, the step of identifying the second codeword contained in the bullet comment, and performing an operation on the bullet comment according to the emotional tendency of the second codeword, includes:

[0023] Matching the bullet comment with the conversion dictionary, determining whether the bullet comment contains a second similar codeword, and if so, inputting the second similar codeword into the financial field event graph to obtain an event related to the second similar codeword;

[0024] Calculating a second semantic similarity of the context of the second similar codeword, and determining that the second similar codeword is a second codeword when the second semantic similarity is higher than a second preset threshold;

[0025] According to the popular semantics of the second codeword in the conversion dictionary, a word-level text sentiment analysis based on the dictionary is performed to obtain the sentiment tendency of the second codeword;

[0026] The emotional tendency of the second codeword is determined. If it is negatively sensitive, codeword masking is performed. If not, an underline is displayed in the barrage during the playback of the barrage. When the video user clicks on the barrage, it is automatically interpreted as the popular semantics of the second codeword in the conversion dictionary.

[0027] Further optionally, in the method described above, the word-level text sentiment analysis method of the dictionary includes calculating the third semantic similarity between the word and the positive and negative polarity seed vocabulary using the synonymy and antonym relationships in the dictionary and the structural hierarchy of the dictionary, and classifying the sentiment tendency of the word according to the size of the third semantic similarity.

[0028] The present invention also provides a system for constructing a causal graph based on code words, the system comprising:

[0029] A financial field affairs graph construction module is used to construct a financial field affairs graph and integrate the financial field affairs graph with the financial field knowledge graph;

[0030] A codeword database and conversion dictionary construction module, used to construct a codeword database and a conversion dictionary matching the codeword database;

[0031] A live content semantic recognition module is used to obtain the semantics of the live content and determine whether it includes the first codeword. If so, the semantics of the live content is input into the financial field event map, and it is inferred whether the phenomenon of video viewers inputting in the barrage occurs later;

[0032] The barrage recognition and execution module is used to recognize the second codeword contained in the barrage and perform operations on the barrage according to the emotional tendency of the second codeword.

[0033] The present invention also provides a computer, comprising a memory and a processor, wherein the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program includes instructions for executing the above method.

[0034] The technical solution provided by the present invention can bring the following beneficial effects:

[0035] The present invention intelligently identifies and analyzes bullet comments containing coded words in live broadcasts in the financial field, can display the popular semantics of the coded bullet comments to viewers in real time, and automatically shields the coded bullet comments containing negative and sensitive content, thereby improving the viewing experience of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The present invention is a flowchart of an embodiment of a method for constructing a causal graph based on code words.

[0037] Figure 2 This is a structural diagram of an embodiment of a system for constructing a code-based logic graph of the present invention.

[0038] Figure 3 This is another structural diagram of an embodiment of the system for constructing a code-based logic graph of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Figure 1 Flow chart of the method of the present invention. Figure 1 As shown, the method for constructing a causal graph based on code words in this embodiment may specifically include the following steps:

[0041] S11, constructing a graph of financial affairs, and integrating the graph of financial affairs with the knowledge graph of financial affairs.

[0042] First, obtain massive amounts of text in the financial field on the Internet and explore the logic behind the massive amounts of text in the financial field.

[0043] Secondly, the logic of the narrative is structured through explicit patterns, and the logic of the narrative from different sources is integrated and abstracted layer by layer to obtain logic of different levels. For the mining of explicit causal logic, a large number of causal event pairs are obtained by manually setting causal patterns. For example, for the sentence "Affected by the Yin-Yang contract incident of Fan, the opening of XX Film and Television and XX Brothers plummeted", a causal event pair such as <Fan's Yin-Yang contract incident led to a sharp drop in the opening of XX Film and Television and XX Brothers> can be structured. By further parsing the cause event and the result event, it can be obtained that the associated subject of the cause event is the person, that is, the film and television star "Fan", and the action of the event is "Yin-Yang contract". The associated subjects in the result event are two media listed companies, Huade Film and Television and XX Brothers, and the action of the event is a sharp drop in stock prices. Combined with the context, the time information of the event can be further clarified, which is June 4, 2018. For such an event pair, it can be further abstracted and generalized into a logical rule, that is, a causal pattern chain such as "Yin-Yang contract of film and television star-> media company stock price drop".

[0044] Furthermore, the logic of things includes causal things, conditional things, reversal things, sequential things, upper and lower things, composition things, and concurrent things; it also includes the emotional polarity of the generalized event action itself, forming an abstract causal pattern chain. Through such a large number of explicit patterns to structure the logic of things, integrate and abstract the knowledge of things from different sources, we can get seven kinds of logic of things at different levels on a large scale, namely causal things, conditional things, reversal things, sequential things, upper and lower things, composition things, and concurrent things. With the help of the end-to-end connection of the knowledge graph, the constructed logic of things is linked to form a graph of things in the financial field. In addition, by generalizing the emotional polarity of the action itself, it can be found that "yin-yang contract" is negative news, and the stock price drop is a negative impact. Therefore, it can be further generalized into a more abstract causal pattern chain of "negative news about celebrities -> bad news about the company".

[0045] Thirdly, the knowledge graph is connected end to end to link the constructed logic and obtain a financial field logic graph based on events.

[0046] Finally, entity recognition and entity linking technology are used to link the entities of events in the financial field event graph to the corresponding financial field knowledge graph, realizing the fusion of the financial field event graph and the financial field knowledge graph. For example, in the event of "Fan's illegal Yin-Yang contract", the character "Fan" can be linked to the character relationship graph containing the character "Fan", such as Sogou's character relationship graph, which led to the stock price decline of media companies such as Guangguang Media, XX Charles Schwab, and XX Brothers, and constructed a fused event graph.

[0047] S12, constructing a codeword database and a conversion dictionary matching the codeword database.

[0048] First, a codeword database is constructed by manually inputting words including codeword meanings.

[0049] Then, the coded words in the coded word library are matched with the first letters of the words in the common dictionary. If the coded words in the coded word library are matched with the words in the common dictionary successfully, the popular semantics of the coded words in the coded word library are determined, such as teller and longan, BT (abnormal), LG (garbage), turkey (money fund, money fund). If the match is unsuccessful, the coded words are semantically matched with the words in the candidate dictionary. When the first letters are not similar, the candidate dictionary is further matched. In this way, a conversion dictionary of popular semantics matched by the coded words is finally obtained. For example, stingy (a listed company that is profitable but does not distribute dividends and increase shares), tax planning (tax avoidance), bricks (ten thousand yuan). Preferably, the candidate dictionary is composed of a negative sentiment dictionary or a network term dictionary.

[0050] Finally, a conversion dictionary of popular semantics matched by the code words in the code word library is constructed.

[0051] S13, obtaining the semantics of the live content and determining whether it includes the first codeword. If so, inputting the semantics of the live content into the financial field event graph, and inferring whether the phenomenon of video viewers inputting in the barrage will occur later.

[0052] First, through speech recognition based on pattern matching, the semantics of the content spoken by the anchor during the live broadcast is converted into the semantics of phrases or sentences. The speech recognition steps based on pattern matching include:

[0053] During the training phase, the user says each word in the vocabulary in turn, and stores its feature vector as a template in the template library.

[0054] In the recognition stage, the feature vector of the input speech is compared with each template in the template library in turn for similarity, and the one with the highest similarity is output as the recognition result.

[0055] Secondly, the semantics of the phrase or sentence is matched with the codeword library to determine whether the semantics of the phrase or sentence contains the first similar codeword. If so, the first similar codeword is input into the financial field event graph to obtain events related to the first similar codeword.

[0056] Next, the first semantic similarity of the context of the first similar codeword is calculated, and when the first semantic similarity is higher than a first preset threshold, the first similar codeword is determined to be the first codeword. Preferably, the first preset threshold is 50%.

[0057] Finally, the semantics of the live content is input into the event graph to infer whether the phenomenon of video viewers inputting in the barrage occurs later. For example, in a live broadcast in the financial field, the semantics of the live content of the anchor saying "I took 10 bricks" is obtained through voice recognition. By matching the words in the conversion dictionary, a similar codeword "brick" is obtained, and then input into the constructed financial field event graph, and the events related to "brick" are obtained as "Li Si took 5 bricks", "Zhang San saved 10 bricks", etc. The first semantic similarity calculation is performed between these events and the recognized "I took 10 bricks.", and the first semantic similarity is 75%, which is higher than the threshold of 50%. It is judged that the similar codeword in "I took 10 bricks." is a codeword, and then "I took 10 bricks." is input into the event graph, and it is inferred that the audience may enter barrages such as "666", "envy", and "awesome" in the future.

[0058] S14, identifying the second codeword contained in the barrage, and performing operations on the barrage according to the emotional tendency of the second codeword.

[0059] First, the barrage is matched with the conversion dictionary to determine whether the barrage contains the second similar codeword. If so, the second similar codeword is input into the financial field event graph to obtain events related to the second similar codeword.

[0060] Secondly, the second semantic similarity of the context of the second similar codeword is calculated, and when the second semantic similarity is higher than a second preset threshold, the second similar codeword is determined to be the second codeword. Preferably, the second preset threshold is 50%.

[0061] Secondly, according to the popular semantics of the second codeword in the conversion dictionary, a dictionary-based word-level text sentiment analysis is performed to obtain the emotional tendency of the second codeword. Among them, the dictionary-based word-level text sentiment analysis method uses the synonymy and antonym relationships in the dictionary and the structural hierarchy of the dictionary to calculate the semantic similarity between the word and the positive and negative polarity seed words, and classifies the sentiment of the word according to the distance of the semantics.

[0062] Finally, the emotional tendency of the second codeword is judged. If it is negative and sensitive (such as politics, pornography, etc.), the codeword is masked. If not, an underline is displayed in the barrage during the playback of the barrage. When the video user clicks on the barrage, it is automatically interpreted as the popular semantics of the second codeword in the conversion dictionary. For example, first, for the barrage "I want to cut my meat and run away", by comparing with the conversion dictionary, "cut meat" is obtained as the second similar codeword; secondly, the relevant events are obtained from the financial field event map, and the second semantic similarity is calculated to be higher than the second preset threshold, and it is determined that the second similar codeword is the second codeword "cut meat". Thirdly, according to the conversion dictionary, the meaning of "cut meat" is "when the loss of a certain investment reaches a predetermined amount, cut the position in time to avoid a greater loss". When the dictionary-based word-level text sentiment analysis is performed and it is found that this codeword is not a negative and sensitive emotional tendency, then during the playback of the barrage, this codeword is underlined, and when the user clicks, it is automatically interpreted as: "When the loss of a certain investment reaches a predetermined amount, cut the position in time to avoid a greater loss."

[0063] The method for constructing a causal graph based on code words provided by the present invention can intelligently identify and analyze bullet comments containing code words in live broadcasts in the financial field, can display the popular semantics of the code word bullet comments to the audience in real time, and automatically shield the code word bullet comments containing negative and sensitive content, thereby improving the user's viewing experience.

[0064] Based on the same inventive concept, another embodiment of the invention provides a system for constructing a logic graph based on code words. Figure 2 As shown in FIG. 1 , it is a structural diagram of a system for constructing a matter-and-reason graph based on code words. The method for constructing a matter-and-reason graph based on code words in this embodiment specifically includes the following modules:

[0065] A financial field affairs graph construction module 21 is used to construct a financial field affairs graph and merge the financial field affairs graph with the financial field knowledge graph;

[0066] A codeword database and conversion dictionary building module 22, used to build a codeword database and a conversion dictionary matching the codeword database;

[0067] The live content semantic recognition module 23 is used to obtain the semantics of the live content and determine whether it includes the first codeword. If so, the semantics of the live content is input into the financial field event map, and it is inferred whether the phenomenon of video viewers inputting in the barrage occurs later;

[0068] The bullet screen recognition and execution module 24 is used to recognize the second codeword contained in the bullet screen, and execute the operation on the bullet screen according to the emotional tendency of the second codeword.

[0069] Based on the same inventive concept, another embodiment of the present invention provides a computer / server, which includes a memory 31 and a processor 32, the memory 31 stores a computer program, the computer program is configured to be executed by the processor 32, and the computer program includes instructions for executing each step in the method of the present invention.

[0070] Parts of the present invention that are not described in detail belong to the common knowledge of those skilled in the art.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit 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 constructing a causal graph based on code words, characterized in that: The method comprises: Constructing a financial field affairs graph, and integrating the financial field affairs graph with the financial field knowledge graph; Constructing a codeword database and a conversion dictionary matching the codeword database; Obtaining the semantics of the live content and determining whether it includes the first codeword, and if so, inputting the semantics of the live content into the integrated financial field event graph, and inferring whether the phenomenon of video viewers inputting in the barrage occurs later; Identify a second codeword contained in the barrage, and perform an operation on the barrage according to the emotional tendency of the second codeword; The step of identifying the second codeword contained in the bullet screen and performing an operation on the bullet screen according to the emotional tendency of the second codeword includes: Matching the bullet comment with the conversion dictionary, determining whether the bullet comment contains a second similar codeword, and if so, inputting the second similar codeword into the financial field event graph to obtain an event related to the second similar codeword; Calculating a second semantic similarity of the context of the second similar codeword, and determining that the second similar codeword is a second codeword when the second semantic similarity is higher than a second preset threshold; According to the popular semantics of the second codeword in the conversion dictionary, a word-level text sentiment analysis based on the dictionary is performed to obtain the sentiment tendency of the second codeword; The emotional tendency of the second codeword is determined. If it is negatively sensitive, codeword masking is performed. If not, an underline is displayed in the barrage during the playback of the barrage. When the video user clicks on the barrage, it is automatically interpreted as the popular semantics of the second codeword in the conversion dictionary.

2. The method according to claim 1, characterized in that: The construction of the financial field theory graph and its integration with the financial field knowledge graph includes: Obtaining massive amounts of text in the financial field on the Internet and mining the logic of the massive amounts of text in the financial field; The logic of things is structured through explicit patterns, and the logic of things from different sources is integrated and abstracted layer by layer to obtain logic of things at different levels; The constructed logic of events is linked by connecting the knowledge graph end to end to obtain the logic graph of the financial field based on events; Entity recognition and entity linking technology are used to link the entities of events in the financial field event graph to the corresponding financial field knowledge graph, thereby realizing the fusion of the financial field event graph and the financial field knowledge graph.

3. The method according to claim 2, characterized in that The logic of affairs includes: cause and effect affairs, conditional affairs, reversal affairs, sequential affairs, superior and subordinate affairs, composition affairs, and concurrent affairs; it also includes the emotional polarity of generalized events and actions themselves, forming an abstract cause and effect pattern chain.

4. The method according to claim 1, characterized in that The step of constructing a codeword database and a conversion dictionary matching the codeword database includes: Manually input words including codeword meanings to build a codeword word library; Matching the first letters of the coded words in the coded word library with the words in the common dictionary, if the coded words in the coded word library match the words in the common dictionary successfully, determining the popular semantics of the coded words in the coded word library, if the match fails, matching the semantics of the coded words with the words in the candidate dictionary; A conversion dictionary of popular semantics matched by the code words in the code word library is constructed.

5. The method according to claim 1, characterized in that The obtaining of the semantics of the live content and determining whether it includes codewords, if so, inputting the semantics of the live content into the financial field event graph, and inferring whether the phenomenon of video viewers inputting in the barrage occurs subsequently, including: Through speech recognition based on pattern matching, the semantics of the content spoken by the anchor during the live broadcast is converted into the semantics of phrases or sentences; Matching the semantics of the phrase or sentence with the codeword library, determining whether the semantics of the phrase or sentence contains a first similar codeword, and if so, inputting the first similar codeword into the financial field event graph to obtain events related to the first similar codeword; Calculating a first semantic similarity of the context of the first similar codeword, and determining that the first similar codeword is a first codeword when the first semantic similarity is higher than a first preset threshold; Input the semantics of the live broadcast content into the event graph to infer whether the phenomenon of video viewers inputting in the barrage will occur later.

6. The method according to claim 5, characterized in that The preset threshold of the first semantic similarity is 50%.

7. The method according to claim 1, characterized in that The word-level text sentiment analysis method of the dictionary includes calculating the third semantic similarity between the word and the positive and negative polarity seed vocabulary by using the synonymy and antonym relationships in the dictionary and the structural hierarchy of the dictionary, and classifying the sentiment tendency of the word according to the size of the third semantic similarity.

8. A system for constructing a causal graph based on code words, characterized in that: The system comprises: A financial field affairs graph construction module is used to construct a financial field affairs graph and integrate the financial field affairs graph with the financial field knowledge graph; A codeword database and conversion dictionary construction module, used to construct a codeword database and a conversion dictionary matching the codeword database; A live content semantic recognition module is used to obtain the semantics of the live content and determine whether it includes the first codeword. If so, the semantics of the live content are input into the integrated financial field event graph, and it is inferred whether the phenomenon of video viewers inputting in the barrage occurs later; A bullet screen recognition and execution module, used to recognize a second codeword contained in the bullet screen, and execute an operation on the bullet screen according to the emotional tendency of the second codeword; Among them, the bullet screen recognition and execution module is specifically used to: Matching the bullet comment with the conversion dictionary, determining whether the bullet comment contains a second similar codeword, and if so, inputting the second similar codeword into the financial field event graph to obtain an event related to the second similar codeword; Calculating a second semantic similarity of the context of the second similar codeword, and determining that the second similar codeword is a second codeword when the second semantic similarity is higher than a second preset threshold; According to the popular semantics of the second codeword in the conversion dictionary, a word-level text sentiment analysis based on the dictionary is performed to obtain the sentiment tendency of the second codeword; The emotional tendency of the second codeword is determined. If it is negatively sensitive, codeword masking is performed. If not, an underline is displayed in the barrage during the playback of the barrage. When the video user clicks on the barrage, it is automatically interpreted as the popular semantics of the second codeword in the conversion dictionary.

9. A computer, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program comprises instructions for executing the method according to any one of claims 1 to 7.

Images