Multi-field-oriented hot event deduction mode mining device and method

Abstract

The invention discloses a multi-field-oriented hot event deduction mode mining device and method, and belongs to the field of natural language processing, and the device comprises a base map construction module, a message recognition module, an event extraction module, a venation sorting module and a mode generation module, the base map construction module constructs a historical deduction mode map of concerned hot events as the basis of mode mining, the message identification module screens out news text materials under concerned topics from massive text materials, and the event extraction module extracts event sentence information in the news text materials; the vein combing module is used for combing related event sentences to form a mode development vein, and the mode generation module is used for mining a hot event deduction mode. According to the method, the development and evolution conditions of the concerned hot events are intuitively displayed, so that mining of the deduction mode of the concerned hot events is realized.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing, and more specifically, to a device and method for mining hot event deduction patterns in multiple fields. Background Art

[0002] As a special form of information, an event occurs at a specific time and place, involves one or more participants, and can usually be described as a change in state. Hot events usually refer to a type of event that has received widespread social attention, aroused great public interest, and had an important impact within a certain period of time. With the rapid development of Internet technology, information has grown at an explosive rate, covering multiple fields including politics, economy, culture, and military. Hot events in these fields not only have a wide impact, but their deductive patterns are also of great significance to government decision-making, corporate strategy formulation, academic research, and public opinion guidance.

[0003] At present, the mining of hot event deduction patterns has become one of the research hotspots, but most of the existing methods and devices focus on data in a single field, so there are many problems. For example, most of these methods focus on the identification of hot events and simple trend analysis, lacking in-depth mining of the internal logic and evolution laws of events. In addition, existing methods and devices often find it difficult to effectively integrate data from different fields, resulting in limited comprehensive understanding and analysis of hot events. At the same time, due to the differences in the characteristics and evolution laws of hot events in different fields, existing methods often lack sufficient versatility and scalability, and are difficult to adapt to diverse application scenarios. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a device and method for mining hot event deduction patterns in multiple fields, which intuitively displays the development and evolution of hot events of concern, thereby realizing the mining of hot event deduction patterns of concern.

[0005] The object of the present invention is achieved through the following solutions:

[0006] A hot event deduction pattern mining device for multiple fields, comprising a base map construction module, a message recognition module, an event extraction module, a context combing module and a pattern generation module;

[0007] The base map construction module is used to construct the historical data of hot events into a historical deduction model as the basic base map for current deduction model mining;

[0008] The message identification module is used to classify the acquired news message materials into different topics and filter out the news message materials under the concerned topics;

[0009] The event extraction module is used to extract events from the news message material of the concerned subject in the form of sentences;

[0010] The context combing module is used to connect the event sentences in time series and abstract them to form a pattern development context;

[0011] The pattern generation module is used to perform node matching and key node analysis on the pattern development context and historical deduction pattern, so as to mine the current deduction pattern.

[0012] Furthermore, the base map construction module specifically adopts an expert knowledge-driven approach to construct a historical deduction model.

[0013] Furthermore, the message identification module specifically uses a K-medoids clustering method to summarize and organize news materials in different fields and under different themes to obtain news message materials under the theme of concern.

[0014] Furthermore, the event extraction module first uses a regularization-based method to divide sentences, and then uses a deep learning BERT model to extract event sentence information.

[0015] Furthermore, the context combing module specifically uses a large text language model to abstract event sentences, constructs a more universal event representation after abstracting the event sentences, and combs the event sentences to form a pattern development context, which is used to subsequently supplement the historical deduction model to form the current deduction model.

[0016] Furthermore, the pattern generation module specifically uses a semantic similarity calculation method to match nodes between the pattern development context and the historical deduction pattern, and uses an in-and-out degree statistics method to perform key node analysis.

[0017] A method for mining hot event deduction patterns in multiple fields, comprising the following steps:

[0018] First, we build a historical deductive model of hot events based on historical data sets, which serves as the base map for subsequent deductive model mining.

[0019] Next, the massive amount of news materials are classified by subject, and news materials with focus topics are selected;

[0020] Next, sentence segmentation and event detection are performed on the news message materials under the topic of interest to obtain event sentences;

[0021] Next, all relevant event sentences are temporally connected and abstracted to obtain the development context of the current model;

[0022] Finally, the pattern development context is matched with the historical deduction pattern and the key nodes are analyzed to obtain the current deduction pattern of the hot events.

[0023] Furthermore, the historical deductive model of the hot events is constructed based on the historical data set as a base map for subsequent deductive model mining, which specifically includes the following sub-steps:

[0024] Step 1: Select historical cases similar or related to the hot events from the historical data set as reference data;

[0025] Step 2: Analyze the data of the selected historical cases, extract key events and changing trends, and construct the key events and changing trends into a basic historical deduction model based on expert knowledge;

[0026] Step 3: Compare and verify the selected historical case data and the basic historical deduction model, modify and improve the unreasonable aspects, and thus obtain the final historical deduction model.

[0027] Furthermore, the subject classification of the massive amount of news message materials and the selection of news message materials of the concerned subject specifically includes the following sub-steps:

[0028] Step 1: Vector set of all news materials Perform clustering of category K and arbitrarily select K vectors as the initial center points medoids;

[0029] Step 2: Assign the remaining vectors to the class represented by the current best medoids according to the principle of being closest to the medoids;

[0030] Step 3: In each class, calculate the sum of the distances between each sample point and other points except the center point, and select the point with the smallest sum of distances as the new medoids;

[0031] Step 4: Repeat the process of step 2-step 3 until all medoids points no longer change or the set maximum number of iterations has been reached;

[0032] Step 5: After steps 1 to 4, all news message materials are divided into K categories, and then news message materials of the category of interest are selected.

[0033] Furthermore, the sentence segmentation and event detection of the news message material under the concerned topic to obtain the event sentence specifically includes the following sub-steps:

[0034] Step 1: Sentence segmentation is performed on the given news article text material, and the sentences ending with a period and containing time elements in the article text are grouped together to form a sentence set X{x1,x2,...,x m};

[0035] Step 2: Select sentence x from the sentence set for event detection;

[0036] Step 3: Perform word segmentation and word embedding preprocessing on the selected sentence x to obtain the mapping vector e of the sentence;

[0037] Step 4: Input the mapping vector e into the trained Bert event sentence detection model and linear classifier to obtain the classification label;

[0038] Step 5: Update the event sentence set. If the classification label is 1, the given sentence x is an event sentence. Add sentence x to the event sentence set V{v1,v2,...,vm}.

[0039] Step 6: Repeat the process of step 2 to step 5 until all sentences in the sentence set are selected. After the selection is completed, the event sentence set V{v1,v2,...,vm} is obtained.

[0040] Furthermore, the above-mentioned sequentially connecting and abstracting all relevant event sentences to obtain the development context of the current model specifically includes the following sub-steps:

[0041] Step 1: Select an event sentence from the event sentence set V{v1,v2,...,vm};

[0042] Step 2: Use the trained time element extraction model to extract time elements from the selected event sentences;

[0043] Step 3: Add the extracted time elements to the time element set to update the time element set;

[0044] Step 4: Repeat the process of step 1 to step 3 until all event sentences in the event sentence set are selected, and then obtain the time element set T{t1, t2, ..., tm};

[0045] Step 5: Based on the time element set T{t1, t2, ..., tm}, connect each event in time sequence to obtain the event development context C{c1, c2, ..., cm};

[0046] Step 6: Based on the abstraction rule, use the text detection model to abstract all nodes c on C;

[0047] Step 7: After abstraction, the pattern development context S{s1,s2,...,sm} is obtained.

[0048] Furthermore, the node matching and key node analysis of the pattern development context and the historical deduction pattern to obtain the current deduction pattern of the hot events of concern specifically includes the following sub-steps:

[0049] Step 1: Select the pre-trained bge semantic similarity calculation model;

[0050] Step 2: Select an abstract pattern from the pattern development context, and input the abstract pattern and the historical deduction pattern into the bge semantic similarity calculation model;

[0051] Step 3: Calculate the semantic similarity between all nodes in the abstract pattern and the historical deductive pattern;

[0052] Step 4: Find the historical deduction pattern node that is most similar to the abstract pattern and obtain the similarity V between them;

[0053] Step 5: If V is greater than the first set value, the abstract pattern is merged into the historical deduction pattern node, otherwise the abstract pattern is added as a new node to the historical deduction pattern;

[0054] Step 6: Repeat the process of step 2 to step 5 until all abstract patterns in the pattern development context are selected, and the basic deductive pattern is obtained after the selection is completed;

[0055] Step 7: Calculate the sum of the in-degree and out-degree of all nodes on the basic deductive model;

[0056] Step 8: Select nodes whose sum of in-degree and out-degree is greater than the second set value as key nodes.

[0057] Furthermore, the first setting value is 0.95; the second setting value is 3.

[0058] Furthermore, after step 8, the following step is also included: using the new model development context to update and expand the current deduction model.

[0059] The beneficial effects of the present invention include:

[0060] (1) The present invention specifically addresses the shortcomings of current hot event deductive pattern mining algorithms, such as poor generalization performance, incomplete pattern consideration, and inaccurate pattern mining, and proposes a multi-domain hot event deductive pattern mining solution. The solution effectively extracts, organizes, and mines text material information containing events through five modules, namely, base map construction, message recognition, event extraction, context combing, and pattern generation, and intuitively displays the development and evolution of hot events, thereby realizing the mining of deductive patterns of hot events.

[0061] (2) The present invention can automatically extract the inherent laws and patterns of hot events from complex news message materials, providing effective support for the prediction and management of hot events.

[0062] (3) The properties and characteristics of hot events of different topics vary. The present invention constructs a base map according to the needs of hot events and then mines based on the base map, which ensures the generalization, scientificity and practicality of the present invention in different fields.

[0063] (4) Most hot event deductive pattern mining algorithms are difficult to fully and accurately reveal the relationships and patterns in hot events. The present invention uses node matching and key node analysis to more accurately and comprehensively mine the deductive patterns of hot events. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 paying creative labor.

[0065] Figure 1 is a general step flow chart of an embodiment of the present invention;

[0066] Figure 2 A flowchart of the specific steps of the base map construction module in an embodiment of the present invention;

[0067] Figure 3 This is a flowchart of the specific steps of the event extraction module in an embodiment of the present invention;

[0068] Figure 4 This is a flowchart of the specific steps of the context combing module in an embodiment of the present invention;

[0069] Figure 5 This is a flowchart of the specific steps of the pattern generation module in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] All features disclosed in all embodiments in this specification, or steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or expanded or replaced in any manner.

[0071] In view of the current situation in the background, in a preferred embodiment, specifically targeting the problems existing in the current hot events deductive pattern mining, the present invention proposes a multi-field hot events deductive pattern mining scheme, including corresponding devices and methods. Among them, it is particularly used to mine cross-chapter news text materials. For example, news text materials in the military and political fields, etc., realize the mining of cross-chapter hot events deductive pattern.

[0072] In the overall concept of the device embodiment of the present invention, as Figure 1As shown, it mainly includes a base map construction module, a message recognition module, an event extraction module, a context combing module and a pattern generation module. Among them, the base map construction module constructs a historical deduction pattern diagram of hot events as the basis for pattern mining, the message recognition module selects news text materials under the focus topic from massive text materials, the event extraction module extracts event sentence information from the news text material, the context combing module combs related event sentences to form a pattern development context, and the pattern generation module realizes the mining of hot event deduction patterns.

[0073] Specifically, in the technical conception of the corresponding module:

[0074] 1) The base map construction module refers to constructing the historical data of hot events into a historical deduction model as the basic base map for mining the current deduction model. The present invention adopts an expert knowledge-driven method to construct the historical deduction model, which can ensure that the constructed historical deduction model is scientific and practical.

[0075] 2) The message identification module is to classify the acquired news message materials into different topics and filter out the news message materials under the topic of interest. For each mining, the amount of acquired news message materials is large and complex. If they are not summarized and sorted, it is difficult to effectively use these data. The present invention uses the K-medoids clustering method to summarize and sort news materials under different fields and different topics, so as to obtain news message materials under the topic of interest.

[0076] 3) The event extraction module is to extract the events in the news message material of the concerned topic in the form of sentences. The present invention first uses a regular expression-based method to divide the sentences, and then uses a deep learning BERT model to extract the event sentence information therein.

[0077] 4) The context combing module refers to connecting all event sentences in time series and abstracting them to form a pattern development context. The present invention uses a large text language model to abstract event sentences. Abstracting event sentences can construct a more universal event representation. Combing event sentences to form a pattern development context facilitates the subsequent supplementation of the historical deduction model to form the current deduction model.

[0078] 5) The pattern generation module is to match the pattern development context and the historical deduction pattern with nodes and analyze the key nodes, so as to mine the current deduction pattern. The present invention uses the method of semantic similarity calculation to match the pattern development context and the historical deduction pattern with nodes, and uses the method of in-out degree statistics to analyze the key nodes.

[0079] In the overall concept of the method embodiment of the present invention, Figure 1As shown, the method includes the following steps: first, a historical deductive model of hot events is constructed based on historical data sets as the base map for subsequent deductive model mining; then, the massive news message materials are subject-classified and news message materials of the focus topic are selected; then, the news message materials under the focus topic are sentence-segmented and event-detected to obtain event sentences; then, all related event sentences are chronologically connected and abstracted to obtain the current pattern development context; finally, the pattern development context is node-matched and the key node analysis is performed on the historical deductive model to obtain the current deductive model of the hot events.

[0080] In a further specific embodiment, the present invention is further described in detail in combination with the specific step flow charts of each module as follows:

[0081] 1) Base map construction module: Expert knowledge and experience are the result of long-term learning, practice and accumulation, and are highly professional and authoritative. In a specific field, experts can accurately analyze and judge complex problems with their rich knowledge and skills. Therefore, this paper adopts an expert knowledge-driven method to construct the historical data set of hot events into a historical deduction model to support the mining of subsequent deduction models. Figure 2 ,The base map construction process driven by expert knowledge includes the following sub-steps:

[0082] Step 1: Select cases similar or related to hot events from historical data sets as reference data;

[0083] Step 2: Analyze the data of the selected historical cases, extract key events and changing trends, and construct the key events and changing trends into a basic historical deduction model based on expert knowledge;

[0084] Step 3: Compare and verify the selected historical case data and the basic historical deduction model, modify and improve the unreasonable aspects, and thus obtain the final historical deduction model.

[0085] 2) Message identification module: The present invention uses the K-medoids clustering algorithm to classify news message materials by topic, and then selects news message materials under the topic of interest. The K-medoids algorithm uses the actual points in the data set as the center points of the cluster, rather than using the average value of all points in the cluster like K-means. The process of using the K-medoids clustering algorithm for topic classification specifically includes the following sub-steps:

[0086] Step 1: Vector set of all news materials Perform clustering of category K and arbitrarily select K vectors as the initial center points medoids;

[0087] Step 2: Assign the remaining vectors to the class represented by the current best medoids according to the principle of being closest to the medoids;

[0088] Step 3: In each class, calculate the sum of the distances between each sample point and other points except the center point, and select the point with the smallest sum of distances as the new medoids;

[0089] Step 4: Repeat the process of step 2-step 3 until all medoids points no longer change or the set maximum number of iterations has been reached;

[0090] After the above steps, all news message materials are divided into K categories, and then the news message materials of the category of interest are selected.

[0091] 3) Event Extraction Module: For news materials of the topic of interest, it is necessary to extract the sentences containing event information from the materials. Figure 3 ,The process of event extraction includes the following sub-steps:

[0092] Step 1: Sentence segmentation is performed on the given news article text material, and the sentences ending with a period and containing time elements in the article text are grouped together to form a sentence set X{x1,x2,...,x m};

[0093] Step 2: Select sentence x from the sentence set for event detection;

[0094] Step 3: Perform preprocessing such as word segmentation and word embedding on the selected sentence x to obtain the mapping vector e of the sentence;

[0095] Step 4: Input the mapping vector e into the trained Bert event sentence detection model and linear classifier to obtain the classification label;

[0096] Step 5: Update the event sentence set. If the classification label is 1, the given sentence x is an event sentence. Add sentence x to the event sentence set V{v1,v2,...,vm}.

[0097] Step 6: Repeat the process of 2-5 until all sentences in the sentence set are selected. After the selection is completed, the event sentence set V{v1,v2,...,vm} is obtained.

[0098] 4) Context Sorting Module: After obtaining the event sentence set of the focus topic, it is necessary to sort it into the form of pattern development context in order to match it with the historical deduction model. Figure 4 , the process of context combing includes the following sub-steps:

[0099] Step 1: Select an event sentence from the event sentence set V{v1,v2,...,vm};

[0100] Step 2: Use the trained time element extraction model to extract time elements from the selected event sentences;

[0101] Step 3: Add the extracted time elements to the time element set to update the time element set;

[0102] Step 4: Repeat the process of 1-3 until all event sentences in the event sentence set are selected, and then obtain the time element set T{t1, t2, ..., tm};

[0103] Step 5: Based on the time element set T{t1, t2, ..., tm}, connect each event in time sequence to obtain the event development context C{c1, c2, ..., cm};

[0104] Step 6: Based on the abstraction rule, use the text detection model to abstract all nodes c on C;

[0105] Step 7: After abstraction, the pattern development context S{s1,s2,...,sm} is obtained.

[0106] 5) Pattern generation module: After obtaining the pattern development context, it is necessary to expand it into the historical deduction pattern and update the historical deduction pattern to obtain the current deduction pattern of the hot events. The process of obtaining the deduction pattern of the hot events is as follows: Figure 5 As shown, it specifically includes the following sub-steps:

[0107] Step 1: Select the pre-trained bge semantic similarity calculation model;

[0108] Step 2: Select an abstract pattern from the pattern development context, and input the abstract pattern and the historical deduction pattern into the bge semantic similarity calculation model;

[0109] Step 3: Calculate the semantic similarity between all nodes in the abstract pattern and the historical deductive pattern;

[0110] Step 4: Find the historical deduction pattern node that is most similar to the abstract pattern and obtain the similarity V between them;

[0111] Step 5: If V is greater than 0.95, merge the abstract pattern into the historical deductive pattern node, otherwise add the abstract pattern as a new node to the historical deductive pattern;

[0112] Step 6: Repeat the process of 2-5 until all abstract patterns in the pattern development context are selected, and the basic deductive pattern is obtained after the selection is completed;

[0113] Step 7: Calculate the sum of the in-degree and out-degree of all nodes on the basic deductive model;

[0114] Step 8: Select nodes whose sum of in-degree and out-degree is greater than 3 as key nodes.

[0115] After completing the above steps, we can obtain the current interpretation model of the hot events, and subsequently we can use the new model development context to update and expand the current interpretation model.

[0116] The units involved in the embodiments of the present invention may be implemented by software or hardware, and the units described may also be arranged in a processor. The names of these units do not, in some cases, limit the units themselves.

[0117] According to one aspect of an embodiment of the present invention, a computer program product or a computer program is provided, the computer program product or the computer program includes a computer instruction, and the computer instruction is stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in the above various optional implementations.

[0118] As another aspect, an embodiment of the present invention further provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment.

Claims

1. A device for mining hot event deduction patterns in multiple fields, characterized by: It includes base map construction module, message recognition module, event extraction module, context combing module and pattern generation module; The base map construction module is used to construct the historical data of hot events into a historical deduction model as the basic base map for current deduction model mining; The message identification module is used to classify the acquired news message materials into different topics and filter out the news message materials under the concerned topics; The event extraction module is used to extract events from the news message material of the concerned subject in the form of sentences; The context combing module is used to connect the event sentences in time series and abstract them to form a pattern development context; The pattern generation module is used to perform node matching and key node analysis on the pattern development context and historical deduction pattern, so as to mine the current deduction pattern.

2. The device for mining hot event deduction patterns in multiple fields according to claim 1 is characterized in that: The base map construction module specifically adopts an expert knowledge-driven approach to construct a historical deduction model.

3. The device for mining hot event deduction patterns in multiple fields according to claim 1 is characterized in that: The message recognition module specifically uses the K-medoids clustering method to summarize and organize news materials in different fields and under different themes to obtain news message materials under the theme of concern.

4. The device for mining hot event deduction patterns in multiple fields according to claim 1 is characterized in that: The event extraction module first uses a regularization-based method to segment sentences, and then uses a deep learning BERT model to extract event sentence information.

5. The device for mining hot event deduction patterns in multiple fields according to claim 1 is characterized in that: The context combing module specifically uses a text large language model to abstract event sentences, constructs a more universal event representation after abstracting the event sentences, and combs the event sentences to form a pattern development context, which is used to subsequently supplement the historical deduction model to form the current deduction model.

6. The device for mining hot event deduction patterns in multiple fields according to claim 1 is characterized in that: The pattern generation module specifically uses a semantic similarity calculation method to match nodes between the pattern development context and the historical deduction pattern, and uses an in-and-out degree statistics method to perform key node analysis.

7. A method for mining hot event deduction patterns in multiple fields, characterized by: The steps include: First, we build a historical deductive model of hot events based on historical data sets, which serves as the base map for subsequent deductive model mining. Next, the massive amount of news materials are classified by subject, and news materials with focus topics are selected; Next, sentence segmentation and event detection are performed on the news message materials under the topic of interest to obtain event sentences; Next, all relevant event sentences are temporally connected and abstracted to obtain the development context of the current model; Finally, the pattern development context is matched with the historical deduction pattern and the key nodes are analyzed to obtain the current deduction pattern of the hot events.

8. The method for mining hot event deduction patterns in multiple fields according to claim 7 is characterized in that: The method of constructing a historical deductive model of a hot event based on a historical data set as a base map for subsequent deductive model mining specifically includes the following sub-steps: Step 1: Select historical cases similar or related to the hot events from the historical data set as reference data; Step 2: Analyze the data of the selected historical cases, extract key events and changing trends, and construct the key events and changing trends into a basic historical deduction model based on expert knowledge; Step 3: Compare and verify the selected historical case data and the basic historical deduction model, modify and improve the unreasonable aspects, and thus obtain the final historical deduction model.

9. The method for mining hot event deduction patterns in multiple fields according to claim 7 is characterized in that: The subject classification of the massive amount of news message materials and the selection of news message materials of the concerned subject specifically includes the following sub-steps: Step 1: Vector set of all news materials Perform clustering of category K and arbitrarily select K vectors as the initial center points medoids; Step 2: Assign the remaining vectors to the class represented by the current best medoids according to the principle of being closest to the medoids; Step 3: In each class, calculate the sum of the distances between each sample point and other points except the center point, and select the point with the smallest sum of distances as the new medoids; Step 4: Repeat the process of step 2-step 3 until all medoids points no longer change or the set maximum number of iterations has been reached; Step 5: After steps 1 to 4, all news message materials are divided into K categories, and then news message materials of the category of interest are selected.

10. The method for mining hot event deduction patterns in multiple fields according to claim 7 is characterized in that: The sentence segmentation and event detection of the news message material under the concerned topic to obtain the event sentence specifically includes the following sub-steps: Step 1: Sentence segmentation is performed on the given news article text material, and the sentences ending with a period and containing time elements in the article text are grouped together to form a sentence set X{x1,x2,...,x m }; Step 2: Select sentence x from the sentence set for event detection; Step 3: Perform word segmentation and word embedding preprocessing on the selected sentence x to obtain the mapping vector e of the sentence; Step 4: Input the mapping vector e into the trained Bert event sentence detection model and linear classifier to obtain the classification label; Step 5: Update the event sentence set. If the classification label is 1, the given sentence x is an event sentence. Add sentence x to the event sentence set V{v1,v2,...,vm}. Step 6: Repeat the process of step 2 to step 5 until all sentences in the sentence set are selected. After the selection is completed, the event sentence set V{v1,v2,...,vm} is obtained.

11. The method for mining hot event deduction patterns in multiple fields according to claim 7 is characterized in that: The above method connects and abstracts all relevant event sentences in time sequence to obtain the development context of the current model, which specifically includes the following sub-steps: Step 1: Select an event sentence from the event sentence set V{v1,v2,...,vm}; Step 2: Use the trained time element extraction model to extract time elements from the selected event sentences; Step 3: Add the extracted time elements to the time element set to update the time element set; Step 4: Repeat the process of step 1 to step 3 until all event sentences in the event sentence set are selected, and then obtain the time element set T{t1, t2, ..., tm}; Step 5: Based on the time element set T{t1, t2, ..., tm}, connect each event in time sequence to obtain the event development context C{c1, c2, ..., cm}; Step 6: Based on the abstraction rule, use the text detection model to abstract all nodes c on C; Step 7: After abstraction, the pattern development context S{s1,s2,...,sm} is obtained.

12. The method for mining hot event deduction patterns in multiple fields according to claim 7 is characterized in that: The node matching and key node analysis of the pattern development context and the historical deduction pattern are performed to obtain the current deduction pattern of the hot events, which specifically includes the following sub-steps: Step 1: Select the pre-trained bge semantic similarity calculation model; Step 2: Select an abstract pattern from the pattern development context, and input the abstract pattern and the historical deduction pattern into the bge semantic similarity calculation model; Step 3: Calculate the semantic similarity between all nodes in the abstract pattern and the historical deductive pattern; Step 4: Find the historical deduction pattern node that is most similar to the abstract pattern and obtain the similarity V between them; Step 5: If V is greater than the first set value, the abstract pattern is merged into the historical deduction pattern node, otherwise the abstract pattern is added as a new node to the historical deduction pattern; Step 6: Repeat the process of step 2 to step 5 until all abstract patterns in the pattern development context are selected, and the basic deductive pattern is obtained after the selection is completed; Step 7: Calculate the sum of the in-degree and out-degree of all nodes on the basic deductive model; Step 8: Select nodes whose sum of in-degree and out-degree is greater than the second set value as key nodes.

13. The method for mining hot event deduction patterns in multiple fields according to claim 12 is characterized in that: The first setting value is 0.95; the second setting value is 3.

14. The method for mining hot event deduction patterns in multiple fields according to claim 12 is characterized in that: After step 8, the following step is also included: using the new model development context to update and expand the current deduction model.

Images