Method and system for detecting factuality of chapter-level event

Abstract

The invention discloses a text-level event factuality detection method and a text-level event factuality detection system. The method comprises the following steps: acquiring a title containing event description and a corpus of a corresponding text; after the corpus is translated, information of main events and key elements is extracted; encoding the extracted bilingual information, and connecting the encoded bilingual information as a global node with the text by adopting an improved model based on a graph attention network; adjusting the weight of the Chinese and English text through grid search; and determining a detection result of event factuality in the title by using the data-enhanced data set. According to the method, key information is obtained through data enhancement before data preprocessing so as to improve the accuracy and efficiency of the model, and expression of different sensitivity degrees of the same text in multiple languages can be obtained through multi-language expression; and the GAT model in the graph model is applied, and the extracted key information is taken as a global node to be combined with the whole text, so that the performance of the model is better, and the robustness is stronger.

Description

Technical Field

[0001] The present invention relates to the technical field of event fact detection, and in particular to a chapter-level event fact detection method and system. Background Art

[0002] The existing chapter-level event fact detection methods have very few data enhancement methods before preprocessing data, and most of them are multilingual representations. Information extraction, as another important task in natural language processing, can also be applied to this work. There have been previous attempts on this basis, introducing the standard-ner package for information extraction tasks, but due to the limitations of accuracy, the improvement effect is not obvious. This results in the key information that can be obtained through data enhancement before data preprocessing. It takes a lot of computing power in the model part to find the representation relationship between text semantics. At the same time, the accuracy of doing so is far less than that of direct extraction, which makes it more difficult to find global structural features and potential feature structures of text.

[0003] Most of the existing chapter-level event fact detection methods focus on how to build models to extract more appropriate semantic features and improve classification performance. They often do not process the data at all or only process it a little before entering the word segmentation module. However, in actual scenarios, it is feasible to enhance the data, which can enable the model to obtain more effective semantic features at the beginning, while reducing the probability of the model obtaining wrong features, reducing the input of noise, and improving the effectiveness of the model. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a chapter-level event fact detection method and system to solve the problem that the existing chapter-level event fact detection method lacks effective data enhancement means in the preprocessing stage, resulting in the model needing to consume a lot of computing power to extract key information and the effect is poor.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for detecting the factuality of events at the chapter level, comprising: obtaining a corpus of titles and corresponding body texts containing event descriptions; extracting information about major events and key elements after translating the corpus; encoding the extracted bilingual information, and connecting the encoded bilingual information with the body text as a global node using an improved model based on a graph attention network; adjusting the weights of Chinese and English texts through a grid search to obtain a data-enhanced data set; and determining the detection result of the factuality of events in the title using the data-enhanced data set.

[0008] As a preferred solution of the chapter-level event fact detection method of the present invention, the step of encoding the extracted bilingual information includes:

[0009] Extracting major events and key elements from an English corpus and a Chinese corpus, wherein the key elements are subjects, verbs, and objects extracted from the major events;

[0010] A segmentation strategy is used to segment texts whose length exceeds a first threshold;

[0011] A Longformer encoder suitable for long text processing is selected to encode the bilingual text that has been pre-extracted and segmented to generate a fixed-length vector representation, and the fixed-length vector representation is input into the improved model based on the graph attention network.

[0012] As a preferred solution of the chapter-level event fact detection method of the present invention, wherein: the use of an improved model based on a graph attention network to connect the encoded bilingual information as a global node with the main text includes:

[0013] Constructing a relationship graph including a main event node, a sentence node, a word node, and a global node into which the key element becomes, and defining connections between the nodes;

[0014] An improved model based on graph attention network is used to update node representation. When each node updates its representation, the weight is reassigned according to the strength of the relationship between it and its neighboring nodes. That is, sentence nodes that are closely related to the global node will be assigned higher weights. The updated global node is used as the first node to add sentences in the text in turn, thereby completing the connection with the main text.

[0015] As a preferred solution of the chapter-level event fact detection method of the present invention, the step of adjusting the weights of Chinese and English texts by grid search to obtain a data-enhanced data set includes:

[0016] Setting a weight ratio representing the importance of the Chinese text relative to the English text, and creating a grid based on the weight ratio, wherein each cell in the grid corresponds to a specific weight combination;

[0017] For each pair of weights in the grid, re-weight the text features using the corresponding ratio and train the model, and measure the performance of the model using a predetermined evaluation criterion on the validation set after each training;

[0018] A weight ratio that optimizes the model performance is selected from all weight combinations, and the weight ratio of the Chinese and English texts is adjusted according to the weight ratio that optimizes the model performance, so as to retain the information of the original English text to the greatest extent.

[0019] As a preferred solution of the chapter-level event fact detection method of the present invention, the detection result of the event fact in the title determined by using the data-enhanced data set includes:

[0020] The nodes and edges in the graph structure are used to capture the complex dependencies in the text, and the influence of key information nodes is strengthened through the attention mechanism to obtain a new vector representation that integrates contextual information.

[0021] The new vector representation of the fused context information is input into the classification layer for prediction. The classification layer maps each node to one of five classification labels, thereby obtaining the detection result of the factuality of the event in the title, wherein the five classification labels are definitely going to happen, may happen, definitely not going to happen, highly unlikely to happen, and unknown.

[0022] As a preferred solution of the chapter-level event fact detection method of the present invention, wherein: using the data-enhanced data set to determine the detection result of the event fact in the title also includes:

[0023] The cross entropy loss function is used to measure the gap between the model prediction value and the true label, and the AdamW optimizer is used to minimize the loss. The model parameters are adjusted through a continuous iterative training process to classify the data-enhanced dataset and determine the detection result of the factuality of the event in the title.

[0024] As a preferred solution of the chapter-level event fact detection method of the present invention, the information of main events and key elements extracted after translating the corpus includes:

[0025] The obtained English corpus containing the title of the event description and the corresponding body text is translated into a Chinese corpus, and the main events and key elements in the English corpus and the Chinese corpus are extracted, and the key elements are the subject, verb and object extracted from the main events.

[0026] In a second aspect, the present invention provides a chapter-level event fact detection system, comprising:

[0027] A data acquisition module is used to obtain a corpus containing event description titles and corresponding body texts;

[0028] An information extraction module, used to extract information about main events and key elements after translating the corpus;

[0029] A multi-dimensional data enhancement module is used to encode the extracted bilingual information, and use an improved model based on graph attention network to connect the encoded bilingual information with the main text as a global node, and adjust the weights of Chinese and English texts through grid search to obtain a data-enhanced dataset;

[0030] The event authenticity recognition module is used to determine the detection result of the event factuality in the title using the data enhanced data set.

[0031] In a third aspect, the present invention provides a computing device, comprising:

[0032] Memory and processor;

[0033] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the chapter-level event factuality detection method are implemented.

[0034] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the chapter-level event factuality detection method.

[0035] Compared with the prior art, the present invention has the following beneficial effects: ① The present invention obtains key information through data enhancement before data preprocessing to improve the accuracy and efficiency of the model. This is the first time that multidimensional data enhancement is applied to chapter-level event fact recognition. Compared with other one-dimensional or non-data enhancement methods, it has a stronger grasp of global structural features and text sensitivity; ② The present invention adopts GPT as the source of multidimensional data enhancement. With the continuous upgrading of technology, it will gain greater advantages in the accuracy and various indicators of data enhancement in the future; ③ The present invention uses a graph model to connect the extracted key nodes as global nodes with the entire article, which not only improves the generalization and accuracy of the model, but also reduces the insensitivity of the model to specific keywords in a certain language by combining Chinese and English. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0037] Figure 1 A schematic diagram of the overall process logic of a method for detecting factuality of a chapter-level event according to an embodiment of the present invention;

[0038] Figure 2 The present invention is a flowchart of the overall process logic of a method for detecting factuality of chapter-level events according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0040] Example 1

[0041] Reference Figure 1-Figure 2 As an embodiment of the present invention, a method for detecting the factuality of a chapter-level event is provided. Figure 1 The specific steps shown include:

[0042] S100: obtaining a corpus containing event description titles and corresponding body texts, translating the corpus and extracting information about major events and key elements;

[0043] S200: Encode the extracted bilingual information, and use an improved model based on graph attention network to connect the encoded bilingual information with the main text as a global node. Adjust the weights of Chinese and English texts through grid search to obtain a data-enhanced dataset.

[0044] S300: Determine a detection result of the factuality of the event in the title using the data-enhanced data set;

[0045] It should be noted that the present invention obtains key information through data enhancement before data preprocessing to improve the accuracy and efficiency of the model. This is the first time that multidimensional data enhancement is applied to chapter-level event fact recognition. Compared with other one-dimensional or non-data enhancement methods, it has a stronger grasp of global structural features and text sensitivity; moreover, the present invention adopts GPT as the source of multidimensional data enhancement. With the continuous upgrading of technology, it will gain greater advantages in the accuracy and various indicators of data enhancement in the future; in addition, the present invention uses a graph model to connect the extracted key nodes as global nodes with the entire article, which not only improves the generalization and accuracy of the model, but also reduces the insensitivity of the model to specific keywords in a certain language by combining Chinese and English.

[0046] In the embodiment of the present application, the above step S100 includes the following sub-steps A1-A2;

[0047] In A1: the obtained English corpus containing the event description title and the corresponding body text is translated into a Chinese corpus;

[0048] In A2: extracting the main events and key elements from the English corpus and the Chinese corpus, where the key elements are the subject, verb and object extracted from the main events;

[0049] In an optional embodiment, the obtained English corpus containing the title of the event description and the corresponding text is translated into a Chinese corpus, and the main events and key elements in the English corpus and the Chinese corpus are extracted, such as Figure 2 The original 2 copies of data are expanded to 6 copies, as shown below:

[0050] The original data is:

[0051] 1) Major events: Voters support for the prime minister;

[0052] 2) English document text;

[0053] The data after data enhancement is:

[0054] 1) Major events: Voters support for the prime minister;

[0055] 2) English document text;

[0056] 3) Key elements: [Voters], [support], [prime minister];

[0057] 4) Main events: Do voters support the prime minister?

[0058] 5) Chinese document text;

[0059] 6) Key elements: [voters], [support], [prime minister];

[0060] It should be noted that bilingual translation uses the original English corpus to form a Chinese corpus through GPT, which expands the size of the corpus; key element extraction extracts the most important elements from the event: subject, verb and object. These elements help the model obtain more effective semantic features at the beginning, while reducing the probability of the model obtaining erroneous features, making the text more sensitive to the events in the model and providing important information for subsequent modules.

[0061] In the embodiment of the present application, the above step S200 includes the following sub-steps B1-B3;

[0062] In B1: encode the extracted bilingual information;

[0063] In B2: an improved model based on graph attention network is used to connect the encoded bilingual information with the main text as a global node;

[0064] In B3: The weights of Chinese and English texts are adjusted through grid search to obtain a data-enhanced dataset;

[0065] Specifically, encoding the extracted bilingual information includes:

[0066] Extracting major events and key elements from English and Chinese corpora;

[0067] A segmentation strategy is used to segment texts whose length exceeds a first threshold;

[0068] The Longformer encoder suitable for long text processing is selected to encode the pre-extracted and segmented bilingual text to generate a fixed-length vector representation, and the fixed-length vector representation is input into the improved model based on the graph attention network;

[0069] It should be noted that the first threshold is 512, because the encoding length of the text needs to be calculated when encoding with long text, and whether it meets the longest length of the selected encoder. Since the most widely used BERT has a maximum encoding length of 512, and half of the corpus length has exceeded this value, if it exceeds this value, the model performance will be seriously degraded. After the experiment used Bert slicing and replaced the encoder, the Longformer encoder with better effect, which is good at processing long text, was selected in this embodiment.

[0070] Specifically, the improved model based on graph attention network is used to connect the encoded bilingual information with the main text as a global node, including:

[0071] Construct a relationship graph containing the main event node, sentence node, word node, and global node where the key element becomes, and define the connection between each node;

[0072] An improved model based on graph attention network is used to update node representation. When updating the representation of each node, the weight is re-assigned according to the strength of the relationship between it and its neighboring nodes. That is, the sentence nodes that are closely related to the global node will be assigned a higher weight. The updated global node is used as the first node to add sentences in the text in turn, thereby completing the connection with the main text.

[0073] It should be noted that constructing a relationship graph containing nodes such as text paragraphs, events, sentences, and key elements means extracting different levels of information units from the encoded vector representation, such as the entire text, specific event descriptions, single sentences, and key elements extracted from sentences (such as subjects, verbs, objects, etc.); then, defining the connection between nodes, for example, the relationship between sentence nodes and event nodes reflects the description of the event in the sentence; the connection between sentence nodes and key element nodes indicates that the sentence contains these key information. In this way, a multi-level graph structure is constructed, which contains various logical connections within the text. The node representation is updated using the graph attention mechanism. The graph attention mechanism allows each node to be assigned different weights according to the strength of the relationship between it and its neighboring nodes when updating its representation. This means that in the graph, each node will recalculate its own feature representation based on the information of the surrounding nodes, and in this calculation process, sentence nodes that are closely related to key information nodes (such as event descriptions, key elements, etc.) will be assigned higher weights, thereby enhancing the model's attention to these key information.

[0074] It should be noted that the widely used GAT is expanded in terms of the graph-based model, and the nodes are divided into main event nodes, sentence nodes, word nodes, and global nodes where key elements become. Compared with the ordinary GAT method, this embodiment adds additional global nodes and builds edges between the global nodes and all sentence nodes, making each sentence more sensitive to the keywords in the main event, reducing the impact of irrelevant noise, and increasing the importance of sentences related to keywords, so that the performance of the model is improved.

[0075] Specifically, the weights of Chinese and English texts are adjusted through grid search to obtain data-enhanced datasets including:

[0076] Set a weight ratio representing the importance of Chinese text relative to English text, and create a grid based on the weight ratio, where each cell in the grid corresponds to a specific weight combination;

[0077] For each pair of weights in the grid, re-weight the text features using the corresponding ratio and train the model, and measure the performance of the model using a predetermined evaluation criterion on the validation set after each training;

[0078] Select the weight ratio that makes the model perform best from all weight combinations, and adjust the weight ratio of Chinese and English texts according to the weight ratio that makes the model perform best, so as to retain the information of the original English text to the greatest extent;

[0079] It should be noted that a reasonable weight range is determined, such as from 0.1 to 0.9, which represents the importance of Chinese text relative to English text. Then, a grid is created based on these weight values, in which each grid corresponds to a specific weight combination. For each pair of weights in the grid, the text features are re-weighted using the corresponding proportions and the model is trained. After each training, the performance of the model is measured using the predetermined evaluation criteria on the validation set, such as accuracy or F1 score. Finally, the weight ratio that makes the model perform best is selected from all weight combinations to maximize the retention of information from the original English text. Once the optimal weights are determined, the generalization ability of the model is verified on an independent test set to ensure the reliability and effectiveness of the model.

[0080] In an optional embodiment, the main event is: Xiao Ming goes fishing today; the key elements are: [Xiao Ming], [fishing], [fish]; the text description is: …………Xiao Hong does not go fishing today. …………

[0081] In the text description, if [Xiao Ming] is not used as a node to be spliced ​​with 'Xiao Hong is not going fishing today', then the simplest Bert model will believe that the probability that Xiao Ming will not go fishing today can reach 70%, because Xiao Hong and Xiao Ming are only one character apart.

[0082] If [Xiao Ming] is used as a node and connected to the sentence using the graph model, the model can fully obtain the information about Xiao Ming, and consider the event "Xiao Hong is not going fishing today" as a useless event. This is an example sentence. For the entire document, [Xiao Ming], [fishing], and [fish] are global key elements and are connected to each node by the graph model. Each node is closely connected to these key elements. Then, these three nodes are directly obtained as global structural features before data preprocessing, and the performance of the model can be improved.

[0083] In the embodiment of the present application, the above step S300 includes the following sub-steps C1-C3;

[0084] In C1: the nodes and edges in the graph structure are used to capture the complex dependencies in the text, the influence of key information nodes is strengthened through the attention mechanism, and a new vector representation that integrates contextual information is obtained;

[0085] In C2: the new vector representation fused with context information is input into the classification layer for prediction. The classification layer maps each node to one of five classification labels, thereby obtaining the detection result of the factuality of the event in the title. The five classification labels are definitely happening (CT+), probably happening (PS+), definitely not happening (CT-), highly unlikely to happen (PS-), and unknown (Uu);

[0086] In C3: the cross entropy loss function is used to measure the gap between the model prediction value and the true label, and the AdamW optimizer is used to minimize the loss. The model parameters are adjusted through continuous iterative training process to classify the data augmented dataset and determine the detection result of the factuality of the event in the title;

[0087] It should be noted that in this embodiment, multidimensional data enhancement is applied to this task through the reference of GPT. By doing a lot of previous work before preprocessing the data, the characteristics of the data are made more sensitive. The multilingual representation can obtain the expression of the same text in multiple languages ​​with different sensitivity levels, and the subject, predicate and object as the most important factors of the main event are extracted as the main elements of the entire text. The GAT model in the graph model is applied to the method provided in this embodiment, so that the extracted key information is combined with the entire text as a global node, making the model have better performance and stronger robustness.

[0088] The above is a schematic scheme of a method for detecting the factuality of chapter-level events in this embodiment. It should be noted that the technical scheme of the system for detecting the factuality of chapter-level events and the technical scheme of the method for detecting the factuality of chapter-level events described above are of the same concept. For details not described in detail in the technical scheme of the system for detecting the factuality of chapter-level events in this embodiment, please refer to the description of the technical scheme of the method for detecting the factuality of chapter-level events described above.

[0089] This embodiment also provides a chapter-level event fact detection system, including:

[0090] A data acquisition module is used to obtain a corpus containing event description titles and corresponding body texts;

[0091] The information extraction module is used to extract information about main events and key elements after translating the corpus;

[0092] The multi-dimensional data enhancement module is used to encode the extracted bilingual information and use an improved model based on graph attention network to connect the encoded bilingual information with the main text as a global node. The weights of Chinese and English texts are adjusted through grid search to obtain a data-enhanced dataset.

[0093] The event authenticity identification module is used to determine the detection results of the event factuality in the title using the data-enhanced dataset.

[0094] The above-mentioned unit modules may be embedded in or independent of the processor in the computer device in the form of hardware, or may be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0095] This embodiment also provides a computing device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for detecting the factuality of a chapter-level event is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0096] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0097] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0098] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0099] Example 2

[0100] Based on the previous embodiment, this embodiment provides an application example of a chapter-level event factuality detection method and system to verify and illustrate the technical effects adopted in this method.

[0101] In this embodiment, the obtained English corpus containing the title of the event description and the corresponding body text is first translated into a Chinese corpus; the main events and key elements in the English corpus and the Chinese corpus are extracted, wherein the key elements are the subject, verb and object extracted from the main events; then the extracted bilingual information is encoded; the improved model based on the graph attention network is used to connect the encoded bilingual information with the body text as a global node; the weights of the Chinese and English texts are adjusted through grid search to obtain a data-enhanced data set; after completing the above submodules, the cross entropy loss function is used as the loss function for classification, and AdamW is selected as the optimizer. The training model obtains five classification results on the test set, including certain occurrence (CT+), possible occurrence (PS+), certain non-occurrence (CT-), high probability non-occurrence (PS-) and unknown (Uu).

[0102] The prediction results are evaluated based on the prediction accuracy of the three main categories (CT+ will definitely occur, CT- will definitely not occur, and PS+ may occur) and the five commonly used macro F1 and micro F1 indicators. The control group is the current better performance and some basic baseline models. The evaluation data is shown in Table 1. Judging from the results, the method provided in this embodiment is the most effective.

[0103] Table 1: Five indicators evaluation data;

[0104]

[0105] From the above, it can be seen that the present invention obtains key information through data enhancement before data preprocessing to improve the accuracy and efficiency of the model. This is the first time that multidimensional data enhancement is applied to chapter-level event fact recognition. Compared with other one-dimensional or non-data enhancement methods, it has a stronger grasp of global structural features and text sensitivity; moreover, the present invention adopts GPT as the source of multidimensional data enhancement. With the continuous upgrading of technology, it will gain greater advantages in the accuracy and various indicators of data enhancement in the future; in addition, the present invention uses a graph model to connect the extracted key nodes as global nodes with the entire article, which not only improves the generalization and accuracy of the model, but also reduces the insensitivity of the model to specific keywords in a certain language by combining Chinese and English.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0107] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0108] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0109] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0111] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for detecting the factuality of a chapter-level event, characterized in that: include: Get a corpus containing event description titles and corresponding body texts; extracting information about main events and key elements after translating the corpus; Encoding the extracted bilingual information, and connecting the encoded bilingual information with the main text as a global node using an improved model based on a graph attention network; Adjust the weights of Chinese and English texts through grid search to obtain a data-enhanced dataset; The data-enhanced dataset is used to determine a detection result of the factuality of the event in the title.

2. The chapter-level event fact detection method according to claim 1, characterized in that: The step of encoding the extracted bilingual information comprises: Extracting major events and key elements from an English corpus and a Chinese corpus, wherein the key elements are subjects, verbs, and objects extracted from the major events; A segmentation strategy is used to segment texts whose length exceeds a first threshold; A Longformer encoder suitable for long text processing is selected to encode the bilingual text that has been pre-extracted and segmented to generate a fixed-length vector representation, and the fixed-length vector representation is input into the improved model based on the graph attention network.

3. The chapter-level event fact detection method according to claim 2, characterized in that: The improved model based on graph attention network is used to connect the encoded bilingual information as a global node with the main text, including: Constructing a relationship graph including a main event node, a sentence node, a word node, and a global node into which the key element becomes, and defining connections between the nodes; An improved model based on graph attention network is used to update node representation. When each node updates its representation, the weight is reassigned according to the strength of the relationship between it and its neighboring nodes. That is, sentence nodes that are closely related to the global node will be assigned higher weights. The updated global node is used as the first node to add sentences in the text in turn, thereby completing the connection with the main text.

4. The chapter-level event fact detection method according to claim 3, characterized in that: The weights of Chinese and English texts are adjusted by grid search to obtain a data-enhanced dataset including: Setting a weight ratio representing the importance of the Chinese text relative to the English text, and creating a grid based on the weight ratio, wherein each cell in the grid corresponds to a specific weight combination; For each pair of weights in the grid, re-weight the text features using the corresponding ratio and train the model, and measure the performance of the model using a predetermined evaluation criterion on the validation set after each training; A weight ratio that optimizes the model performance is selected from all weight combinations, and the weight ratio of the Chinese and English texts is adjusted according to the weight ratio that optimizes the model performance, so as to retain the information of the original English text to the greatest extent.

5. The chapter-level event fact detection method according to claim 4, characterized in that: The detection results of determining the factuality of events in titles using the data-enhanced dataset include: The nodes and edges in the graph structure are used to capture the complex dependencies in the text, and the influence of key information nodes is strengthened through the attention mechanism to obtain a new vector representation that integrates contextual information. The new vector representation of the fused context information is input into the classification layer for prediction. The classification layer maps each node to one of five classification labels, thereby obtaining the detection result of the factuality of the event in the title, wherein the five classification labels are definitely going to happen, may happen, definitely not going to happen, highly unlikely to happen, and unknown.

6. The method for detecting factuality of chapter-level events as claimed in claim 5, characterized in that: The detection results of determining the factuality of the event in the title using the data-enhanced dataset also include: The cross entropy loss function is used to measure the gap between the model prediction value and the true label, and the AdamW optimizer is used to minimize the loss. The model parameters are adjusted through a continuous iterative training process to classify the data-enhanced dataset and determine the detection result of the factuality of the event in the title.

7. The method for detecting factuality of chapter-level events as claimed in claim 1, characterized in that: After translating the corpus, the information of main events and key elements extracted includes: The obtained English corpus containing the title of the event description and the corresponding body text is translated into a Chinese corpus, and the main events and key elements in the English corpus and the Chinese corpus are extracted, and the key elements are the subject, verb and object extracted from the main events.

8. A system using the chapter-level event fact detection method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to obtain a corpus containing event description titles and corresponding body texts; An information extraction module, used to extract information about main events and key elements after translating the corpus; A multi-dimensional data enhancement module is used to encode the extracted bilingual information, and use an improved model based on graph attention network to connect the encoded bilingual information with the main text as a global node, and adjust the weights of Chinese and English texts through grid search to obtain a data-enhanced dataset; The event authenticity recognition module is used to determine the detection result of the event factuality in the title using the data enhanced data set.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the chapter-level event factuality detection method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the chapter-level event factuality detection method described in any one of claims 1 to 7.

Images