Description mining method and device, equipment, medium and program product

Abstract

The embodiment of the invention discloses an argument mining method and device, equipment, a medium and a program product, and the method comprises the steps: obtaining a clause set of event news data, and dividing clauses belonging to speech in the clause set into publisher content and speech content according to a preset speech trigger word; performing publisher three-element extraction on the content of each publisher, and combining each extracted publisher three-element with the corresponding speech to determine a speech to be processed; performing period division on the to-be-processed speech, and determining the to-be-processed speech belonging to the same period as the to-be-processed speech in the corresponding period; and performing argument extraction and feature word extraction on the to-be-processed speech in each period through a pre-constructed argument generation prompt and a pre-trained feature word extraction model, splicing each obtained core argument and each main feature word, and determining period argument information corresponding to the period. The data processing amount required in the argument mining process is reduced, and the accuracy and efficiency of argument mining are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an argument mining method, device, equipment, medium and program product. Background Art

[0002] With the advent of the information age, especially in today's digital age, information is growing explosively, and people can obtain massive amounts of text data from different channels. These text data can come from many fields such as news media, academic research, and social media. People urgently need to quickly extract valuable opinions from these massive amounts of information to meet the needs of knowledge acquisition and decision-making support. At the same time, with the rapid development and changes of society, the understanding and grasp of different opinions has become crucial, which has further promoted the development of argument mining technology in information.

[0003] At present, when processing hot events, especially data from sources such as news media, it mainly relies on traditional public opinion analysis models, relying on manual statistical analysis and sentiment analysis techniques. However, under the above processing methods, only relatively shallow information in the hot event data can be analyzed, and it is difficult to obtain the core arguments of the hot events. It is also more susceptible to human subjective factors, which reduces the accuracy of the analysis results. Even with the development of natural language processing (NLP) technology, artificial intelligence analysis technology based on NLP technology can be used to assist in the analysis and extraction of text semantics and intrinsic information. However, NLP technology, which is commonly used for processing various types of text information, often needs to analyze and process the entire input text data, and is not sensitive to time characteristics and the special structure of text data. It is difficult to efficiently and accurately analyze and process the data corresponding to hot events. Summary of the invention

[0004] The present invention provides a method, device, equipment, medium and program product for mining arguments, which pre-divide news data according to its special structure, extract the required information in a targeted manner, ensure that the data used for extracting and determining arguments contains complete speech information, and reduce the amount of data calculation in the process of extracting effective information. At the same time, the time characteristics between different speeches are fully considered to divide them, so that the arguments in the event news data can be efficiently and accurately mined and determined based on the time characteristics.

[0005] In a first aspect, an embodiment of the present invention provides an argument mining method, comprising:

[0006] Obtain a sentence set corresponding to the event news data, and divide each sentence belonging to the speech in the sentence set into publisher content and speech content according to the preset speech trigger words;

[0007] Extract the three elements of each publisher's content respectively, and combine each extracted three elements of the publisher with the corresponding speech to determine a speech to be processed;

[0008] Divide each of the pending speeches into periods, and determine the pending speeches belonging to the same period as the pending speeches in the period corresponding to the period;

[0009] Through the pre-constructed argument generation prompts and pre-trained feature word extraction model, argument extraction and feature word extraction are performed on the speeches to be processed in each period corresponding to each period, and the obtained core arguments and main feature words are spliced ​​to determine the periodic argument information corresponding to the period.

[0010] In a second aspect, an embodiment of the present invention further provides an argument mining device, comprising:

[0011] A content acquisition module is used to acquire a set of sentences corresponding to the event news data, and divide each sentence belonging to the speech in the set of sentences into publisher content and speech content according to a preset speech trigger word;

[0012] The speech determination module is used to extract the three elements of the publisher from the content of each publisher, and combine each extracted three elements of the publisher with the corresponding speech to determine a speech to be processed;

[0013] A periodic speech determination module is used to divide each to-be-processed speech into periods, and determine the to-be-processed speeches belonging to the same period as the to-be-processed speeches within the period corresponding to the period;

[0014] The argument information determination module is used to extract arguments and feature words from the speeches to be processed in each period corresponding to each period through pre-built argument generation prompts and pre-trained feature word extraction models, and to splice the obtained core arguments and main feature words to determine the period argument information corresponding to the period.

[0015] In a third aspect, an embodiment of the present invention further provides an argument mining device, comprising:

[0016] at least one processor; and

[0017] a memory communicatively coupled to the at least one processor;

[0018] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can execute the argument mining method provided by an embodiment of the present invention.

[0019] In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to execute the argument mining method provided by the embodiment of the present invention.

[0020] In a fifth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, is used to execute the argument mining method provided by an embodiment of the present invention.

[0021] The embodiments of the present invention provide an argument mining method, device, equipment, medium and program product, which obtains a sentence set corresponding to event news data, and divides each sentence belonging to a speech in the sentence set into publisher content and speech content according to preset speech trigger words; extracts the publisher's three elements from each publisher's content respectively, and combines each extracted publisher's three elements with the corresponding speech to determine a speech to be processed; divides each speech to be processed into periods, and determines the speech to be processed belonging to the same period as the speech to be processed in the period corresponding to the period; extracts arguments and feature words from the speech to be processed in each period corresponding to each period through a pre-constructed argument generation prompt and a pre-trained feature word extraction model, splices the obtained core arguments and main feature words, and determines the period argument information corresponding to the period. By adopting the above technical solution, when it is necessary to mine arguments for event news data, the sentences corresponding to the event news data will first be screened and divided according to the preset speech trigger words, and only the information as the publisher content part in the sentence belonging to the speech will be extracted from the publisher's three elements, so that the speech to be processed finally constructed contains all speech information related to the argument mining needs, and reduces the amount of data calculation in the speech construction process. Then, by dividing each speech to be processed into periods, the timeliness and time evolution characteristics of the event news data are fully considered, and the arguments and feature words of the speeches to be processed belonging to the same period are generated based on the period, and the internal connection between the speeches to be processed with the same time characteristics is fully considered. The periodic argument information corresponding to the period is determined, which reduces the amount of data processing required in the argument mining process and improves the accuracy and efficiency of argument mining.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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.

[0024] Figure 1 A flowchart of an argument mining method provided in Embodiment 1 of the present invention;

[0025] Figure 2 A flowchart of an argument mining method provided in Embodiment 2 of the present invention;

[0026] Figure 3 A flowchart of an argument mining method provided in Embodiment 3 of the present invention;

[0027] Figure 4 A flowchart of an argument mining method provided in Embodiment 4 of the present invention;

[0028] Figure 5 A schematic diagram of the structure of an argument mining device provided in Embodiment 5 of the present invention;

[0029] Figure 6 A schematic diagram of the structure of an argument mining device provided in Example 6 of the present invention. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] Embodiment 1

[0033] Figure 1 This is a flow chart of a method for argument mining provided in the first embodiment of the present invention. The embodiment of the present invention can be applied to extracting arguments from a large amount of text information generated in hot news events. The method can be executed by an argument mining device, which can be implemented by software and / or hardware, and the argument mining device can be configured in an argument mining device. Optionally, the argument mining device can be a notebook, a desktop computer, a smart tablet, a programmable device, etc., which is not limited by the embodiment of the present invention.

[0034] like Figure 1 As shown, an argument mining method provided by an embodiment of the present invention specifically includes the following steps:

[0035] S101. Obtain a sentence set corresponding to event news data, and divide each sentence belonging to speech in the sentence set into publisher content and speech content according to preset speech trigger words.

[0036] In this embodiment, the event news data can be specifically understood as a collection of news type data obtained from multiple different sources based on an event. The sentence set can be specifically understood as a collection of multiple sentences obtained by dividing the news content in the event news data into sentences based on text language logic. Exemplarily, the news content in the event news data can be divided based on specific symbols, such as ".", "!", "?", etc., to obtain multiple sentences, and the set of each sentence is determined as a sentence set corresponding to the event news data. The preset speech trigger words can be specifically understood as words pre-set according to actual conditions to determine whether a sentence contains speech content. Exemplarily, a speech trigger word library containing a speech trigger word list can be pre-set, and the words in the speech trigger word list can be used as preset speech trigger words. The preset speech trigger words can also be determined by other methods, and the embodiment of the present invention does not limit this. The publisher content can be specifically understood as content related to the region, organization and person where the news is published. The speech content can be specifically understood as content related to the specific information that the news is published to express.

[0037] Specifically, when certain hot events occur, multiple news reports on the hot events will be made by different channels. All the acquired news reports that need to be mined can be determined as event news data, and the event news data can be divided into sentences based on text language logic to obtain a set of sentences corresponding to the event news data. Then, for each sentence in the sentence set, it is determined whether each sentence belongs to speech through the preset speech trigger words, and the sentences determined to belong to speech are divided according to the preset speech trigger words to obtain the publisher content containing information related to the news publisher, and the speech content related to the specific information required to be expressed by the news.

[0038] In the embodiment of the present invention, by setting preset speech trigger words to screen the sentence set, the data required for subsequent argument mining are all complete speech information that contains both publisher information and speech content. At the same time, based on the preset speech trigger words, the sentences belonging to the speech are divided, and the expression structure of the news type data is fully utilized to quickly and completely separate the information related to the news publisher in the sentence, thereby reducing the amount of data calculation required for extracting effective information from event news data.

[0039] S102, extracting the three elements of each publisher from the content of each publisher, and combining each extracted three elements of the publisher with the corresponding speech content to determine a speech to be processed.

[0040] In this embodiment, the three elements of the publisher can be specifically understood as the three basic elements related to the news publisher that need to be considered when publishing information in the field of news dissemination. For example, the three elements of the publisher may include region, organization and person, and the embodiment of the present invention does not limit this. The speech to be processed can be specifically understood as the speech that needs to be mined for arguments.

[0041] Specifically, for the publisher content in each sentence, the three publisher elements contained therein are extracted to obtain at least one of the three publisher elements, and then the three publisher elements extracted from the sentence are concatenated with the speech content obtained by dividing the sentence to obtain the speech to be processed corresponding to the sentence.

[0042] It can be understood that the above processing process is a loop execution process, that is, it traverses all the sentences belonging to the speech and completes the determination of the pending speech of each sentence belonging to the speech.

[0043] S103, dividing the to-be-processed speeches into periods, and determining the to-be-processed speeches belonging to the same period as the to-be-processed speeches within the period corresponding to the period.

[0044] Specifically, the speeches to be processed are divided into periods according to the time information contained in each speech to be processed, that is, the release time of each speech to be processed in the same period is similar. After the period division of all speeches to be processed is completed, the speeches to be processed in each period are determined as the speeches to be processed in the period corresponding to the period.

[0045] It can be understood that the cycle used when dividing the speech to be processed into periods can be obtained by repeatedly dividing the optimal period according to the number of speeches to be processed contained in different time periods, or the cycle length can be pre-set according to actual conditions, and each speech to be processed can be directly divided according to the pre-set cycle length. The embodiments of the present invention are not limited to this.

[0046] S104. The arguments and feature words of the speeches to be processed in each period corresponding to each period are extracted through the pre-constructed argument generation prompts and the pre-trained feature word extraction model, and the obtained core arguments and main feature words are spliced ​​to determine the period argument information corresponding to the period.

[0047] In this embodiment, the argument generation prompt can be specifically understood as using natural language to guide or stimulate the artificial intelligence model, so that the artificial intelligence model can complete the artificial intelligence prompt word (prompt) for extracting argument information contained in the input content. Among them, when constructing the prompt, it is necessary to follow certain answer specifications, and the answer specification can be "Please state the key information and provide a concise and clear reply that conforms to objective facts based on the user's inquiry, and return it in the corresponding format."

[0048] Exemplarily, the pre-constructed argument generation prompt in an embodiment of the present invention may be "You are an AI assistant that is good at natural language processing and can extract core ideas from text. Please analyze the core arguments of a given speech based on [relevant examples]."

[0049] Among them, [relevant example] can be expressed as:

[0050] Comments: The visual effects of this movie are very stunning, but the plot is a bit cliché.

[0051] Core point: The visual effects are stunning, but the plot is old-fashioned.

[0052] That is, the output method of the pre-constructed argument generation prompt in the embodiment of the present invention may be:

[0053] Please analyze the core arguments of the following statements:

[0054] Speech: {{input_text}}

[0055] Core arguments:

[0056] In this embodiment, the pre-trained feature word extraction model can be specifically understood as a neural network model that is pre-trained according to actual conditions and is used to extract and classify the features of the text information input therein, and then output the words related to the core viewpoints in the text information as feature words. The periodic argument information can be specifically understood as a combination of information related to the same argument within the period.

[0057] Specifically, after clarifying all the pending speeches in each cycle corresponding to each cycle, the pending speeches in each cycle can be aggregated and screened first, that is, the pending speeches in each cycle that may belong to the same argument can be aggregated for the purpose of extracting relevant information about the same argument, and then for each possible argument, the pre-constructed argument generation prompt and the pre-trained feature word extraction model are used to extract argument extraction and feature word extraction for the pending speeches in each cycle belonging to the same argument, and the extracted core arguments are concatenated with the main feature words to obtain the periodic argument information corresponding to the core argument.

[0058] It is understandable that for the speeches to be processed in each cycle corresponding to the same argument, a core argument can be extracted, as well as multiple different main feature words given to the core argument in the speeches to be processed in each cycle. Therefore, when completing the splicing of the cycle argument information, multiple main feature words belonging to the speeches to be processed in different cycles but belonging to the same core argument are spliced ​​with the core argument to obtain the cycle argument information corresponding to the core argument. At the same time, there can be multiple core arguments in a cycle, that is, there can be multiple cycle argument information corresponding to the cycle, and the number is the same as the number of core arguments.

[0059] The technical solution of this embodiment is to obtain a sentence set corresponding to event news data, and divide each sentence belonging to the speech in the sentence set into publisher content and speech content according to preset speech trigger words; extract the publisher's three elements from each publisher's content respectively, and combine each extracted publisher's three elements with the corresponding speech to determine a speech to be processed; divide the speeches to be processed into periods, and determine the speeches to be processed belonging to the same period as the speeches to be processed in the period corresponding to the period; extract arguments and feature words from the speeches to be processed in each period corresponding to each period through pre-constructed argument generation prompts and pre-trained feature word extraction models, splice the obtained core arguments and main feature words, and determine the period argument information corresponding to the period. By adopting the above technical solution, when it is necessary to mine arguments for event news data, the sentences corresponding to the event news data will first be screened and divided according to the preset speech trigger words, and only the information as the publisher content part in the sentence belonging to the speech will be extracted from the publisher's three elements, so that the speech to be processed finally constructed contains all speech information related to the argument mining needs, and reduces the amount of data calculation in the speech construction process. Then, by dividing each speech to be processed into periods, the timeliness and time evolution characteristics of the event news data are fully considered, and the arguments and feature words of the speeches to be processed belonging to the same period are generated based on the period, and the internal connection between the speeches to be processed with the same time characteristics is fully considered. The periodic argument information corresponding to the period is determined, which reduces the amount of data processing required in the argument mining process and improves the accuracy and efficiency of argument mining.

[0060] Embodiment 2

[0061] Figure 2 The flowchart of a method for argument mining provided in the second embodiment of the present invention is further optimized on the basis of the above-mentioned optional technical solutions. According to the preset speech trigger words, the slot segmentation technology is introduced into the process of processing each sentence, and each sentence belonging to the speech is divided into the corresponding publisher content and speech content, so that only the part of the sentence belonging to the publisher content needs to be extracted based on the entity recognition algorithm technology, which is related to the specificity of the news type data, so that the speech to be processed finally constructed contains all speech information related to the argument mining needs. According to the time information contained in each speech to be processed, each speech to be processed is aggregated and divided into different time spans, and the most appropriate time span is selected as the basis for periodic division, fully considering the influence of time on the evolution of the arguments corresponding to the event news, so that the internal connection between each speech to be processed with the same time characteristics can be fully reflected in the periodic argument information, reducing the amount of data processing required in the argument mining process, and improving the accuracy and efficiency of argument mining.

[0062] like Figure 2 As shown, a method for mining arguments provided by Embodiment 2 of the present invention specifically includes the following steps:

[0063] S201. Obtain a sentence set corresponding to event news data.

[0064] S202. For each sentence in the sentence set, determine whether the sentence contains a preset speech trigger word; if so, execute S204; if not, execute S203.

[0065] Specifically, the sentence set is traversed, that is, for each sentence in the sentence set, the sentence is segmented to obtain a segmentation set corresponding to the sentence, and then each word in the segmentation set is compared with the preset speech trigger word to determine whether the segmentation set contains the same segmentation as the preset speech trigger word. As long as there is any segmentation in the segmentation set that is the same as the preset speech trigger word, S204 is executed; if all segmentations do not overlap with the preset speech trigger word, S203 is executed.

[0066] S203. Determine that the clause is not a clause that belongs to speech.

[0067] Specifically, when it is determined that the word set corresponding to the sentence does not contain the preset speech trigger word, it can be considered that the content in the sentence does not belong to speech and there is no need to extract the argument. Therefore, the sentence can be determined as a sentence that does not belong to speech, and the sentence will not be processed in the subsequent argument mining process.

[0068] S204: Determine that the sentence is a sentence belonging to speech, and segment the sentence into slots according to preset speech trigger words to obtain publisher content and speech content corresponding to the sentence.

[0069] Specifically, when it is determined that the word set corresponding to the sentence contains a preset speech trigger word, the content in the sentence can be considered to belong to speech and the argument extraction for the speech in the sentence needs to be completed in the subsequent processing process. In order to effectively extract the information related to the composition of speech contained in the sentence, based on the special expression structure of event news type data, the speech trigger word can be preset as the basis for division, and the sentence can be divided by combining the slot segmentation method to obtain the publisher content containing information related to the news publisher and the speech content related to the specific expression information of the news.

[0070] S205, extracting the three elements of each publisher from the content of each publisher respectively, and combining each extracted three elements of the publisher with the corresponding speech content to determine a speech to be processed.

[0071] S206: Aggregate and divide the speeches to be processed into different time spans according to the time information contained in the speeches to be processed.

[0072] Specifically, each speech to be processed should contain time information related to its publication time, and then each speech to be processed can be aggregated and divided into different time spans, such as day, week, and month, to obtain the aggregation results of the corresponding time span aggregation.

[0073] For example, assuming that each pending speech is an event that lasted for three months, and the speeches obtained by processing news about the event were released at different time points, then when the pending speeches are aggregated and divided based on the time span of days, the speeches about the event news published on the same day can be aggregated into one group, and a maximum of ninety groups of pending speech groups can be obtained.

[0074] S207, determining the time span that meets the preset period division conditions as the target period, and dividing each speech to be processed into periods according to the target period and the time information contained in each speech to be processed, and determining the speeches to be processed belonging to the same period as the speeches to be processed within the period corresponding to the period.

[0075] In this embodiment, the preset period division condition can be specifically understood as a condition predetermined according to actual conditions to ensure that there are sufficient numbers of speeches for argument extraction in each period.

[0076] Specifically, after completing an aggregation division according to a time span, the group of pending speech obtained by the aggregation division is compared with the preset period division condition. If the condition is met, the time span of the aggregation division is determined as the target period, and the period division of each pending speech is completed according to the target period and the time information contained in each pending speech. If the condition is not met, it can be considered that the time span is not suitable for the period division of each pending speech. At this time, a new time span will be used to re-aggregate and divide each pending speech, that is, return to the step of aggregating and dividing each pending speech according to the time span, and compare the group of pending speech obtained by aggregation with the preset period division condition again until the condition is met. Then, the pending speeches belonging to the same period can be determined as the pending speeches in the period corresponding to the period.

[0077] Exemplarily, each speech to be processed may first be aggregated and divided by day according to the time information contained therein, and it is determined whether the number of days after the division is greater than 50, and whether the proportion of the speech group to be processed with a total amount of more than 20 speeches in the speech group to be processed corresponding to each day is greater than 10% of all the speech groups to be processed. If all of them are satisfied, the target period may be determined as "day"; if any of them is not satisfied, each speech to be processed may be aggregated and divided by week according to the time information contained therein, and it is determined whether the number of weeks after the division is greater than 50, and whether the proportion of the speech group to be processed with a total amount of more than 20 speeches in the speech group to be processed corresponding to each week is greater than 10% of all the speech groups to be processed. If all of them are satisfied, the target period may be determined as "week"; if any of them is not satisfied, each speech to be processed may be aggregated and divided by month according to the time information contained therein, and it is determined whether the proportion of the speech group to be processed with a total amount of more than 100 speeches in the speech group to be processed corresponding to each month is greater than 10% of all the speech groups to be processed. If so, the target period may be determined as "month"; if not, the target period may be determined as "year".

[0078] It is understandable that the total amount of speech and the speech proportion conditions in the above examples can be adjusted according to actual conditions, and the embodiments of the present invention do not limit this.

[0079] Optionally, before dividing each speech to be processed into periods, each speech to be processed may be deduplicated first to reduce the amount of data required to be processed for period division.

[0080] S208. For each cycle corresponding to the speeches to be processed in each cycle, vectorize the speeches to be processed in each cycle to determine the corresponding cycle speech vectors.

[0081] Specifically, for the speeches to be processed in each cycle corresponding to each cycle, in order to clarify the semantic information contained therein, the vectorization model can be called to perform an Embedding operation on the speeches to be processed in each cycle to obtain the cycle speech vectors corresponding to the speeches to be processed in each cycle.

[0082] S209: Cluster and screen the speech vectors in each period to determine the target speech cluster corresponding to the period.

[0083] In this embodiment, the target speech cluster can be specifically understood as a cluster that is obtained by clustering and screening the speech vectors in each period based on semantic information, and that belongs to the same argument and has a large number of individuals.

[0084] Specifically, the speech vectors in each period are first clustered so that the speeches in each period belonging to the same argument can be clustered into clusters based on semantic features. In order to avoid too many arguments in the same period due to insufficient clustering, each cluster can be screened according to the number of individuals contained in different clusters, and only the clusters with a larger number of individuals are retained as the target speech clusters corresponding to the period, to ensure that the core arguments in the same period can be accurately extracted.

[0085] Optionally, clustering and screening are performed on speech vectors in each period to determine the target speech cluster corresponding to the period, which can be specifically achieved in the following ways:

[0086] 1) Cluster the speech vectors in each period according to semantic features to determine at least one speech cluster to be screened.

[0087] 2) Screening each speech cluster to be screened according to the number of speeches contained in each speech cluster to be screened, and determining the speech cluster to be screened with the number of speeches within the first preset number as the target speech cluster corresponding to the period.

[0088] In this embodiment, the preset number can be specifically understood as the number of arguments that need to be extracted within one cycle and is pre-set according to actual conditions.

[0089] Specifically, an unsupervised clustering algorithm is used to cluster the speech vectors in each period according to semantic features, so that the speech vectors in the same cluster obtained by clustering have similar arguments, and each speech cluster obtained by clustering is determined as a speech cluster to be screened. Then, each speech cluster to be screened is sorted from high to low according to the number of speech vectors in the period contained therein, and the speech cluster to be screened whose number of speech vectors in the period is in the front of a preset number is considered to be the target speech cluster that needs to be retained in this period for extracting arguments. Exemplarily, the preset number can be 8, and the embodiment of the present invention does not limit this.

[0090] S210, extracting arguments from each target speech cluster through pre-constructed argument generation prompts, and determining the core arguments of each target speech cluster.

[0091] S211. Extract feature words from each target speech cluster using a pre-trained feature word extraction model to determine the main feature words of each target speech cluster.

[0092] It is understandable that S210 and S211 may be executed simultaneously or in any order, and the embodiment of the present invention is not limited to this.

[0093] S212, concatenating the core arguments and main characteristic words belonging to the same target speech cluster to determine the speech cluster argument information of the target speech cluster.

[0094] Specifically, since the same target speech cluster can determine a core argument after being processed by the pre-constructed argument generation prompt, and the speech vectors in each period of the target speech cluster may have different main feature words extracted from the speech vectors in different periods after feature word extraction by the pre-trained feature word extraction model, multiple main feature words can be extracted. Then, each main feature word and the core argument can be spliced ​​according to the association relationship, and the spliced ​​content is determined as the speech cluster argument information of the target speech cluster.

[0095] S213. Combine the speech cluster argument information of each target speech cluster within the period to determine the period argument information corresponding to the period.

[0096] The technical solution of this embodiment introduces slot segmentation technology into the process of processing each sentence based on preset speech trigger words, and divides each sentence belonging to the speech into corresponding publisher content and speech content, so that only the publisher content part of the sentence needs to be extracted based on entity recognition algorithm technology, which is related to the specificity of news type data, so that the speech to be processed finally constructed contains all speech information related to the argument mining needs. According to the time information contained in each speech to be processed, each speech to be processed is aggregated and divided into different time spans, and the most appropriate time span is selected as the basis for period division, fully considering the impact of time on the evolution of arguments corresponding to event news, so that the internal connection between each speech to be processed with the same time characteristics can be fully reflected in the periodic argument information, reducing the amount of data processing required in the argument mining process, and improving the accuracy and efficiency of argument mining.

[0097] Embodiment 3

[0098] Figure 3 A flow chart of an argument mining method provided for the third embodiment of the present invention. The technical solution of the embodiment of the present invention is further optimized on the basis of the above-mentioned optional technical solutions. After determining the periodic argument information corresponding to each period, that is, extracting the arguments from the event news data based on the period, the relationship between the periodic argument information of different periods is mined, and when it is determined that there is a preliminary content association relationship between two adjacent periods, the evolutionary relationship of the arguments between adjacent periods is comprehensively determined based on the intersection probability of the main feature words between different periods and the semantic similarity of the arguments between different periods, thereby realizing automatic analysis of the evolution of event news arguments based on time, improving the efficiency, accuracy and objectivity of mining the potential relationships of arguments between different periods, and improving the adaptability of the obtained analysis results to the event news analysis.

[0099] like Figure 3 As shown, an argument analysis method provided by an embodiment of the present invention specifically includes the following steps:

[0100] S301, obtaining a sentence set corresponding to the event news data, and dividing each sentence belonging to the speech in the sentence set into publisher content and speech content according to a preset speech trigger word.

[0101] S302: extract the three elements of each publisher's content respectively, and combine each extracted three elements of the publisher with the corresponding speech content to determine a speech to be processed.

[0102] S303: Divide the to-be-processed speeches into periods, and determine the to-be-processed speeches belonging to the same period as the to-be-processed speeches within the period corresponding to the period.

[0103] S304. The arguments and feature words of the speeches to be processed in each period corresponding to each period are extracted through the pre-constructed argument generation prompts and the pre-trained feature word extraction model, and the obtained core arguments and main feature words are spliced ​​to determine the period argument information corresponding to the period.

[0104] It can be understood that S301-S304 have been described in detail in the above embodiments, and will not be described in detail in the embodiment of the present invention.

[0105] S305. Traverse each period in chronological order, and mine the content association relationship between the periodic argument information in the first period and the periodic argument information in the second period through pre-constructed argument relationship mining prompts.

[0106] The first cycle is the previous cycle of the second cycle.

[0107] In this embodiment, the argument relationship mining prompt can be specifically understood as a prompt that uses natural language to guide or stimulate the artificial intelligence model so that the artificial intelligence model can complete the mining of the content relevance between two sets of input periodic argument information.

[0108] For example, the pre-built argument relationship mining prompt in the embodiment of the present invention may be "You are an AI assistant that is good at analyzing and extracting information. The task is to find an argument that has the greatest correlation with a specific central argument from a given argument list and provide a brief reason." A specific related example is given below:

[0109] Central argument: Remote work improves employee productivity

[0110] List of arguments:

[0111] [“Remote work leads to communication inconvenience”,

[0112] "Flexible working hours improve employee satisfaction",

[0113] "Telecommuting saves commuting time",

[0114] “Office rent is a major expense for businesses”]

[0115] That is, the output method of the pre-constructed argument relationship mining prompt in the embodiment of the present invention can be:

[0116] Now, please analyze the following central argument and list of arguments and give your conclusion.

[0117] Central argument: {{s.get('argument')}}

[0118] Argument list: {{aList}}

[0119] Your output should look like this:

[0120] {

[0121] "name":"argument name or content",

[0122] "reason":"A brief reason for choosing this argument"

[0123] }

[0124] Specifically, after the periodic argument information of all periods is determined, since each period has corresponding time information, each period can be traversed according to the periodic time sequence from early to late, and content association relationships can be mined in groups of two. The earlier period in each group is determined as the first period, and the later period is determined as the second period. The argument information of each period in the first period and the argument information of each period in the second period are input according to the pre-constructed argument relationship mining prompts to complete the mining of content association relationships between the two.

[0125] Exemplarily, assuming that the event news data is processed to obtain periodic argument information corresponding to four periods, the first two periods of the four periods can be processed as the first period and the second period respectively, and then the second and third periods of the four periods can be processed as the first period and the second period respectively, and finally the last two periods of the four periods can be processed as the first period and the second period respectively, so as to realize the content association relationship mining operation using the pre-constructed argument relationship mining prompt in the embodiment of the present invention. At the same time, since different periods can include multiple periodic argument information at the same time, when processing a group of first and second periods, the periodic argument information in the first period can be firstly mined with the periodic argument information in the second period in turn through the pre-constructed argument relationship mining prompt until all the periodic argument information in each period and each period are traversed.

[0126] S306. For two periodic argument information having a content association relationship, the periodic argument information belonging to the first period is determined as the first periodic argument information, and the periodic argument information belonging to the second period is determined as the second periodic argument information.

[0127] Specifically, after mining the content association relationship between periodic argument information belonging to different periods based on the pre-constructed argument mining prompts, the subsequent evolution relationship will only be determined for the two periodic argument information with content association relationship. At this time, for any two periodic argument information with content association relationship, the periodic argument information belonging to the first period can be determined as the first periodic argument information, and the periodic argument information belonging to the second period can be determined as the second periodic argument information.

[0128] S307. Determine a similarity weight based on the semantic similarity between the first cycle argument information and the second cycle argument information.

[0129] Specifically, the semantic similarity between the first cycle argument information and the second cycle argument information is determined through vector space models, machine learning models, semantic embedding, knowledge graphs, dependency syntax analysis, semantic role labeling or other semantic similarity calculation tools, and the semantic similarity is determined as the similarity weight between the first cycle argument information and the second cycle argument information.

[0130] S308. Determine the intersection probability based on the feature word intersection of the main feature words in the first cycle argument information and the main feature words in the second cycle argument information.

[0131] Specifically, since both the first cycle argument information and the second cycle argument information contain multiple main feature words, it can be considered that the first cycle argument information contains a first main feature word set, and the second cycle argument information contains a second main feature word set. The intersection of the two can be obtained by finding the intersection of the feature words of the two cycle argument information, and then the ratio of the feature word intersection to the feature word union of the two cycle argument information can be determined as the intersection probability of the first cycle argument information and the second cycle argument information.

[0132] It is understandable that S307 and S308 may be executed simultaneously or in any order, and the embodiment of the present invention is not limited to this.

[0133] S309, perform weighted decision judgment on the similarity weight and the intersection probability, and determine the argument evolution relationship between the first cycle argument information and the second cycle argument information based on the obtained judgment result and the preset judgment threshold.

[0134] In this embodiment, the preset determination threshold may be specifically understood as a value that is preset according to actual conditions and is used to determine whether there is a clear evolutionary relationship between different periodic argument information.

[0135] Optionally, the argument evolution relationship between the first cycle argument information and the second cycle argument information is determined according to the obtained determination result and a preset determination threshold, which can be specifically implemented in the following manner:

[0136] 1) If the judgment result is greater than the preset judgment threshold, it is determined that there is a correlation between the first cycle argument information and the second cycle argument information, and the evolution reasons of the first cycle argument information and the second cycle argument information are determined, and the correlation and the evolution reasons are combined to determine the argument evolution relationship between the first cycle argument information and the second cycle argument information.

[0137] 2) If the determination result is less than or equal to the preset determination threshold, the evolution relationship between the first cycle argument information and the second cycle argument information is determined.

[0138] In this embodiment, the reasons for the evolution of things depend on his understanding of the content and correlation between the information of different periodic arguments, and the reasons for the evolution are inferred through the context understanding ability.

[0139] Specifically, when the judgment result is greater than the preset judgment threshold, it can be considered that there is a correlation between the first-cycle argument information and the second-cycle argument information, and then the evolution reasons between the first-cycle argument information and the second-cycle argument information can be determined based on the context understanding ability output of the large model. The argument evolution relationship between the first-cycle argument information and the second-cycle argument information can be obtained by splicing the two together; when the judgment result is less than or equal to the preset judgment threshold, it can be considered that although there is a certain content connection between the first-cycle argument information and the second-cycle argument information, after deep deduction and understanding, it is determined that there is no sufficient argument evolution relationship between the two. At this time, the argument evolution relationship between the first-cycle argument information and the second-cycle argument information is determined.

[0140] Optionally, after determining the argument evolution relationship between the first period argument information and the second period argument information according to the obtained determination result and the preset determination threshold, the method further includes:

[0141] According to the evolutionary relationship of each argument, construct the overall argument evolutionary relationship or argument reasoning map.

[0142] In this embodiment, the overall argument evolution relationship can be specifically understood as connecting the argument evolution relationships of all periods corresponding to the event news data in series to obtain the evolution relationship of all arguments in the event news on the overall timeline.

[0143] In this embodiment, the argument and reason graph can be specifically understood as connecting a certain argument in the event news data in series according to the periodic time sequence corresponding to the evolutionary relationship of each argument, and fusing the evolutionary causes of the cause and reason of the adjacent periods to obtain the overall argument and reason evolution cause, and then adding the overall argument and reason evolution cause to the graph obtained in the series of argument evolution relationships.

[0144] The technical solution of this embodiment, after determining the periodic argument information corresponding to each period, that is, after mining and extracting arguments from event news data based on periods, mines the relationship between periodic argument information of different periods, and when it is determined that there is a preliminary content association relationship between two adjacent periods, the evolutionary relationship of arguments between adjacent periods is comprehensively determined based on the intersection probability of main feature words between different periods and the semantic similarity of arguments between different periods, thereby realizing automatic analysis of the evolution of event news arguments based on time, improving the efficiency, accuracy and objectivity of mining potential relationships between arguments between different periods, and improving the adaptability of the obtained analysis results to event news analysis.

[0145] Embodiment 4

[0146] Figure 4 A flow chart of an argument mining method provided for Embodiment 4 of the present invention. The technical solution of the embodiment of the present invention is further optimized on the basis of the above-mentioned optional technical solutions. After determining the periodic argument information corresponding to each period, that is, extracting the argument mining of event news data according to the period, the remarks published by different publishers in the same period are aggregated, so that the core of the remarks put forward by different publishers in one period can be extracted and confirmed, thereby realizing the comparison between the views of different publishers in the same period, and determining the evolution of the views of the same publisher in different time periods during the fermentation process of event news, thereby improving the effectiveness of event news analysis.

[0147] like Figure 4 As shown, an argument mining method provided by an embodiment of the present invention specifically includes the following steps:

[0148] S401, obtaining a sentence set corresponding to the event news data, and dividing each sentence belonging to the speech in the sentence set into publisher content and speech content according to a preset speech trigger word.

[0149] S402: extract the three elements of each publisher's content respectively, and combine each extracted three elements of the publisher with the corresponding speech content to determine a speech to be processed.

[0150] S403: Divide the to-be-processed speeches into periods, and determine the to-be-processed speeches belonging to the same period as the to-be-processed speeches within the period corresponding to the period.

[0151] S404, extracting arguments and feature words from the speeches to be processed in each period corresponding to each period through the pre-constructed argument generation prompts and the pre-trained feature word extraction model, concatenating the obtained core arguments and main feature words, and determining the period argument information corresponding to the period.

[0152] It can be understood that S401-S404 have been described in detail in the above embodiments, and will not be described in detail in the embodiment of the present invention.

[0153] S405. For each period, perform publisher aggregation on the period argument information corresponding to the period, and determine at least one publisher argument information group corresponding to the period.

[0154] Specifically, since the periodic argument information corresponding to each period contains the three elements of the publisher, it must contain information related to the publisher. Therefore, for the periodic argument information in each period, the periodic argument information corresponding to each period can be aggregated according to the different publishers corresponding to the periodic argument information, and the periodic argument information belonging to the same publisher can be combined to determine a publisher argument information group.

[0155] S406. Perform core opinion mining on each publisher's argument information group through pre-constructed core opinion mining prompts to determine the core opinions of the publishers corresponding to the publisher's argument information group within the period.

[0156] In this embodiment, the core idea mining prompt can be specifically understood as using natural language to guide or stimulate the artificial intelligence model, so that the artificial intelligence model can complete the prompt for core idea mining for the input argument information belonging to the same publisher.

[0157] For example, the pre-constructed core opinion mining prompt in the embodiment of the present invention may be "You are an AI assistant that is good at text analysis and opinion extraction. The task is to analyze the given speech and extract the potential main opinions from it. You need to give a brief reason of no more than 15 words to support your opinion extraction." A specific relevant example is given below:

[0158] Comments: Although the price of this product is relatively high, the quality and service are excellent.

[0159] Output:

[0160] {

[0161] "Underlying opinion": "High quality products deserve higher prices",

[0162] "Reason":"Recognize the price of high-value products"

[0163] }

[0164] Please note:

[0165] 1. The potential viewpoint extracted must accurately reflect the deeper meaning of the speech;

[0166] 2. The output JSON format must be correct;

[0167] 3. The reason should be concise and clear, not exceeding 15 words;

[0168] 4. If there is no obvious underlying opinion in the speech, then output {"Underlying opinion":"No obvious underlying opinion","Reason":""}.

[0169] Now, please analyze the following statements and give your conclusions.

[0170] Statement: {{input_statement}}

[0171] Specifically, after publisher aggregation for each period, multiple publisher argument information groups belonging to different publishers can be obtained. In order to clarify the core viewpoints of the information released by the same publisher in the period, the publisher argument information group belonging to a publisher can be input through pre-built core viewpoint mining prompts, and the output can be determined as the core viewpoint of the publisher corresponding to the publisher argument information group in the corresponding period.

[0172] S407, comparing the core opinions of the publishers in the same period, and determining the opinion comparison results within the period.

[0173] Specifically, by presenting the core viewpoints of different publishers in the same period in a preset format, the comparison between the core viewpoints of different publishers can be achieved, and the presentation result can be determined as the comparison result of the viewpoints within the period.

[0174] S408. Compare the core opinions of the same publisher in different cycles to determine the publisher's opinion evolution results.

[0175] Specifically, since the opinions of the same publisher may change as the event news develops, multiple core opinions of the same publisher can be extracted from different periods, and each core opinion can be presented in a periodic time sequence according to a pre-set format, reflecting the evolution of the publisher's core opinions over time, and obtaining the corresponding publisher's opinion evolution results.

[0176] It is understandable that S407 and S408 may be executed simultaneously or in any order, and the embodiment of the present invention is not limited to this.

[0177] The technical solution of this embodiment, after determining the periodic argument information corresponding to each period, that is, after mining and extracting the arguments of the event news data according to the period, aggregates the remarks published by different publishers in the same period, so that the core of the remarks put forward by different publishers in one period can be extracted and confirmed, thereby realizing the comparison between the views of different publishers in the same period, and determining the evolution of the views of the same publisher in different time periods during the fermentation process of the event news, thereby improving the effectiveness of the event news analysis.

[0178] Embodiment 5

[0179] Figure 5 A schematic diagram of the structure of an argument mining device provided in Embodiment 5 of the present invention is shown in FIG. Figure 5 As shown, the argument mining device includes: a content acquisition module 51, a speech determination module 52, a periodic speech determination module 53 and an argument information determination module 54.

[0180] Among them, the content acquisition module 51 is used to obtain a sentence set corresponding to the event news data, and divide each sentence belonging to the speech in the sentence set into publisher content and speech content according to the preset speech trigger words; the speech determination module 52 is used to extract the publisher's three elements for each publisher's content respectively, and combine each extracted publisher's three elements with the corresponding speech to determine a speech to be processed; the periodic speech determination module 53 is used to divide the speeches to be processed into periods, and determine the speeches to be processed belonging to the same period as the speeches to be processed in the period corresponding to the period; the argument information determination module 54 is used to extract arguments and feature words for the speeches to be processed in each period corresponding to each period through pre-constructed argument generation prompts and pre-trained feature word extraction models, splice the obtained core arguments and main feature words, and determine the periodic argument information corresponding to the period.

[0181] The technical solution of this embodiment, when it is necessary to mine the arguments of event news data, will first screen and divide the sentences corresponding to the event news data according to the preset speech trigger words, and only extract the three elements of the publisher from the information in the sentences belonging to the speech as the content part of the publisher, so that the speech to be processed finally constructed contains all the speech information related to the argument mining needs, and reduces the amount of data calculation in the speech construction process. Then, by dividing the speeches to be processed into periods, the timeliness and time evolution characteristics of the event news data are fully considered, and the arguments and feature words are extracted for the speeches to be processed belonging to the same period based on the period, and the internal connection between the speeches to be processed with the same time characteristics is fully considered to complete the determination of the periodic argument information corresponding to the period, which reduces the amount of data processing required in the argument mining process and improves the accuracy and efficiency of argument mining.

[0182] Optionally, the content acquisition module 51 is specifically used to: obtain a sentence set corresponding to the event news data, for each sentence in the sentence set, if the sentence includes a preset speech trigger word, determine the sentence as a sentence belonging to the speech, and perform slot segmentation on the sentence according to the preset speech trigger word to obtain the publisher content and speech content corresponding to the sentence; if the sentence does not include the preset speech trigger word, determine the sentence as a sentence that does not belong to the speech.

[0183] Optionally, the periodic speech determination module 53 is specifically used to: aggregate and divide each speech to be processed into different time spans according to the time information contained in each speech to be processed; determine the time span that meets the preset period division conditions as the target period, and divide each speech to be processed into periods according to the target period and the time information contained in each speech to be processed; determine the speeches to be processed belonging to the same period as the speeches to be processed within the period corresponding to the period.

[0184] Optionally, the argument information determination module 54 is specifically used to:

[0185] For each cycle corresponding to the speech to be processed in each cycle, the speech to be processed in each cycle is vectorized to determine the corresponding cycle speech vector;

[0186] Cluster and screen the speech vectors in each period to determine the target speech cluster corresponding to the period;

[0187] Extract arguments from each target speech cluster through pre-constructed argument generation prompts to determine the core arguments of each target speech cluster;

[0188] Extract feature words from each target speech cluster through a pre-trained feature word extraction model to determine the main feature words of each target speech cluster;

[0189] The core arguments and main characteristic words belonging to the same target speech cluster are spliced ​​together to determine the speech cluster argument information of the target speech cluster;

[0190] The speech cluster argument information of each target speech cluster within the period is combined to determine the period argument information corresponding to the period.

[0191] Optionally, clustering and screening the speech vectors in each period to determine the target speech cluster corresponding to the period includes:

[0192] Clustering the speech vectors in each period according to semantic features to determine at least one speech cluster to be screened;

[0193] Each speech cluster to be screened is screened according to the number of speeches contained in each speech cluster to be screened, and the speech cluster to be screened whose number of speeches is within a preset number is determined as the target speech cluster corresponding to the period.

[0194] Optionally, the argument mining device further includes: an argument evolution relationship determination module, which is used to:

[0195] After determining the periodic argument information corresponding to the period, traverse each period in the periodic time sequence, and mine the content association relationship between the periodic argument information in the first period and the periodic argument information in the second period through the pre-constructed argument relationship mining prompt; wherein the first period is the previous period of the second period;

[0196] For two periodic argument information having a content association relationship, the periodic argument information belonging to the first period is determined as the first periodic argument information, and the periodic argument information belonging to the second period is determined as the second periodic argument information;

[0197] Determine a similarity weight according to the semantic similarity between the first cycle argument information and the second cycle argument information;

[0198] Determine the intersection probability according to the intersection of the main characteristic words in the first cycle argument information and the main characteristic words in the second cycle argument information;

[0199] The similarity weights and intersection probabilities are weighted for decision making, and the argument evolution relationship between the first cycle argument information and the second cycle argument information is determined based on the obtained decision results and the preset decision threshold.

[0200] Optionally, determining the argument evolution relationship between the first period argument information and the second period argument information according to the obtained determination result and a preset determination threshold value includes:

[0201] If the determination result is greater than the preset determination threshold, it is determined that there is a correlation between the first cycle argument information and the second cycle argument information, and the cause of the evolution of the first cycle argument information and the second cycle argument information is determined, and the existence of the correlation and the cause of the evolution of the matter are combined to determine the argument evolution relationship between the first cycle argument information and the second cycle argument information;

[0202] If the determination result is less than or equal to the preset determination threshold, the evolution relationship between the first cycle argument information and the second cycle argument information is determined.

[0203] Optionally, after determining the argument evolution relationship between the first period argument information and the second period argument information according to the obtained determination result and the preset determination threshold, the method further includes:

[0204] According to the evolutionary relationship of each argument, construct the overall argument evolutionary relationship or argument reasoning map.

[0205] Optionally, the argument mining device further includes: an argument comparison module, which is used to:

[0206] After determining the periodic argument information corresponding to the period, for each period, the periodic argument information corresponding to the period is aggregated by publishers to determine at least one publisher argument information group corresponding to the period;

[0207] Conduct core opinion mining on each publisher's argument information group through pre-built core opinion mining prompts to determine the core opinions of the publishers corresponding to the publisher's argument information group within the period;

[0208] Compare the core opinions of each publisher in the same period and determine the opinion comparison results within the period;

[0209] Compare the core opinions of the same publisher in different cycles to determine the evolution of the publisher's opinions.

[0210] The argument mining device provided in the embodiment of the present invention can execute the argument mining method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0211] Embodiment 6

[0212] Figure 6 A schematic diagram of the structure of an argument mining device provided for Embodiment 6 of the present invention. The argument mining device 60 may be intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The argument mining device 60 may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0213] like Figure 6 As shown, the argument mining device 60 includes at least one processor 61, and a memory connected to the at least one processor 61, such as a read-only memory (ROM) 62, a random access memory (RAM) 63, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 to the random access memory (RAM) 63. In the RAM 63, various programs and data required for the operation of the argument mining device 60 can also be stored. The processor 61, the ROM 62 and the RAM 63 are connected to each other through a bus 64. The input / output (I / O) interface 65 is also connected to the bus 64.

[0214] Multiple components in the argument mining device 60 are connected to the I / O interface 65, including: an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a disk, an optical disk, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the argument mining device 60 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0215] The processor 61 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 61 executes the various methods and processes described above, such as the argument mining method.

[0216] In some embodiments, the argument mining method may be implemented as a computer program that is tangibly contained in a computer-readable storage medium, such as a storage unit 68. In some embodiments, part or all of the computer program may be loaded and / or installed on the computing power distribution device 60 via the ROM 62 and / or the communication unit 69. When the computer program is loaded into the RAM 63 and executed by the processor 61, one or more steps of the argument mining method described above may be performed. Alternatively, in other embodiments, the processor 61 may be configured to execute the argument mining method in any other appropriate manner (e.g., by means of firmware).

[0217] Optionally, an embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the argument mining method provided by any embodiment of the present invention.

[0218] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0219] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0220] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0221] In order to provide interaction with the user, the systems and techniques described herein can be implemented on a computing power distribution device, which has: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball), through which the user can provide input to the computing power distribution device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0222] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0223] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0224] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0225] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for argument mining, characterized in that: include: Acquire a sentence set corresponding to the event news data, and divide each sentence belonging to speech in the sentence set into publisher content and speech content according to a preset speech trigger word; Extracting the publisher's three elements from each publisher's content respectively, and combining each extracted publisher's three elements with the corresponding speech content to determine a speech to be processed; Dividing the to-be-processed speeches into periods, and determining the to-be-processed speeches belonging to the same period as the to-be-processed speeches within the period corresponding to the period; Through the pre-constructed argument generation prompts and pre-trained feature word extraction model, argument extraction and feature word extraction are performed on the speeches to be processed in each period corresponding to each period, and the obtained core arguments and main feature words are spliced ​​to determine the period argument information corresponding to the period.

2. The argument mining method according to claim 1, characterized in that: The step of dividing each sentence belonging to speech in the sentence set into publisher content and speech content according to the preset speech trigger word includes: For each clause in the clause set, If the sentence includes a preset speech trigger word, the sentence is determined to be a sentence belonging to speech, and the sentence is slot-segmented according to the preset speech trigger word to obtain publisher content and speech content corresponding to the sentence; If the sentence does not include a preset speech trigger word, the sentence is determined to be a sentence that does not belong to a speech.

3. The argument mining method according to claim 1, characterized in that: The periodic division of each of the to-be-processed speeches includes: According to the time information contained in each of the to-be-processed speeches, the to-be-processed speeches are aggregated and divided into different time spans; A time span satisfying a preset period division condition is determined as a target period, and each of the to-be-processed speeches is divided into periods according to the target period and the time information contained in each of the to-be-processed speeches.

4. The argument mining method according to claim 1, characterized in that: The pre-built argument generation prompt and pre-trained feature word extraction model are used to extract arguments and feature words from the speeches to be processed in each cycle corresponding to each cycle, and the obtained core arguments and main feature words are spliced ​​to determine the periodic argument information corresponding to the cycle, including: For each cycle corresponding to each speech to be processed in the cycle, vectorization processing is performed on the speech to be processed in each cycle to determine the corresponding cycle speech vector; Clustering and screening the speech vectors in each of the periods to determine the target speech cluster corresponding to the period; Extracting arguments from each of the target speech clusters through pre-constructed argument generation prompts to determine the core arguments of each of the target speech clusters; Extracting feature words from each of the target speech clusters using a pre-trained feature word extraction model to determine the main feature words of each of the target speech clusters; splicing the core arguments and main characteristic words belonging to the same target speech cluster to determine the speech cluster argument information of the target speech cluster; The speech cluster argument information of each of the target speech clusters within the period is combined to determine the period argument information corresponding to the period.

5. The argument mining method according to claim 4, characterized in that: The clustering and screening of speech vectors in each of the periods to determine a target speech cluster corresponding to the period includes: Clustering the speech vectors in each of the periods according to semantic features to determine at least one speech cluster to be screened; Each of the speech clusters to be screened is screened according to the number of speeches contained in each of the speech clusters to be screened, and the speech cluster to be screened whose number of speeches is before a preset number is determined as the target speech cluster corresponding to the period.

6. The argument mining method according to claim 1, characterized in that: After determining the period argument information corresponding to the period, the method further comprises: Traversing each of the periods in order of period time, mining the content association relationship between the periodic argument information in the first period and the periodic argument information in the second period through pre-constructed argument relationship mining prompts; wherein the first period is the previous period of the second period; For two periodic argument information having a content association relationship, the periodic argument information belonging to the first period is determined as the first periodic argument information, and the periodic argument information belonging to the second period is determined as the second periodic argument information; Determine a similarity weight based on the semantic similarity between the first cycle argument information and the second cycle argument information; Determine the intersection probability according to the intersection of the main characteristic words in the first cycle argument information and the main characteristic words in the second cycle argument information; The similarity weight and the intersection probability are subjected to weighted decision making, and the argument evolution relationship between the first cycle argument information and the second cycle argument information is determined based on the obtained decision result and a preset decision threshold.

7. The argument mining method according to claim 6, characterized in that: The step of determining the argument evolution relationship between the first period argument information and the second period argument information according to the obtained determination result and the preset determination threshold value includes: If the determination result is greater than a preset determination threshold, it is determined that there is a correlation between the first cycle argument information and the second cycle argument information, and the cause of the evolution of the first cycle argument information and the second cycle argument information is determined, and the correlation and the cause of the evolution are combined to determine the argument evolution relationship between the first cycle argument information and the second cycle argument information; If the determination result is less than or equal to a preset determination threshold, the evolution relationship between the first cycle argument information and the second cycle argument information is determined regardless of the point of evolution.

8. The argument mining method according to claim 7, characterized in that: After determining the argument evolution relationship between the first period argument information and the second period argument information according to the obtained determination result and the preset determination threshold, the method further includes: According to the evolutionary relationships of each argument, an overall argument evolutionary relationship or an argument reasoning map is constructed.

9. The argument mining method according to claim 1, characterized in that: After determining the period argument information corresponding to the period, the method further comprises: For each period, performing publisher aggregation on the period argument information corresponding to the period, and determining at least one publisher argument information group corresponding to the period; Performing core opinion mining on each publisher's argument information group through pre-constructed core opinion mining prompts to determine the core opinions of the publisher corresponding to the publisher's argument information group in the period; Compare the core opinions of each publisher in the same period and determine the opinion comparison results within the period; Compare the core opinions of the same publisher in different cycles to determine the evolution of the publisher's opinions.

10. An argument mining device, characterized in that: include: A content acquisition module, used to acquire a set of sentences corresponding to event news data, and divide each sentence belonging to speech in the set of sentences into publisher content and speech content according to preset speech trigger words; The speech determination module is used to extract the publisher's three elements from each publisher's content, and combine each extracted publisher's three elements with the corresponding speech to determine a speech to be processed; A periodic speech determination module, used for dividing the to-be-processed speeches into periods, and determining the to-be-processed speeches belonging to the same period as the to-be-processed speeches within the period corresponding to the period; The argument information determination module is used to extract arguments and feature words from the speeches to be processed in each period corresponding to each period through pre-built argument generation prompts and pre-trained feature word extraction models, and to splice the obtained core arguments and main feature words to determine the period argument information corresponding to the period.

11. An argument mining device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; Wherein, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the argument mining method described in any one of claims 1-9.

12. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions are used to perform the argument mining method as described in any one of claims 1-9 when executed by a computer processor.

13. A computer program product, comprising a computer program, which, when executed by a processor, implements the argument mining method as described in any one of claims 1 to 9.

Images