Event detection method and device fusing multi-source information
By integrating multi-source information into the event detection method, and utilizing candidate trigger word extraction models and trigger word extraction models, the problem of inaccurate detection in the context of single-source information is solved, and more accurate event detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FUDAN UNIVERSITY
- Filing Date
- 2024-12-30
- Publication Date
- 2026-05-01
AI Technical Summary
Existing event detection methods are mainly based on single-source information, which leads to inaccurate detection results. They lack the ability to integrate multi-source information and cannot comprehensively detect events.
By integrating multi-source information into an event detection method, and utilizing a candidate trigger word extraction model and a trigger word extraction model, multiple texts are preprocessed, matched, generated, and classified to produce accurate event types.
It achieves effective fusion of multi-source information, generates more accurate event detection results, and improves the effectiveness of trigger word extraction and event classification.
Smart Images
Figure CN120011555B_ABST
Abstract
Description
Event detection method and device that integrates multi-source information Technical Field
[0001] This invention relates to an event detection method, specifically to an event detection method and apparatus that integrates multi-source information. Background Technology
[0002] With the development of the internet and the economy, people can access a vast amount of news information from various fields online. The relevant event information within these news articles is crucial for researchers in these fields. Therefore, information extraction tasks such as event detection play a vital role in the news event domain. As a subtask of event extraction, event detection aims to extract event trigger words and their corresponding event types from unstructured language text. Event trigger words are words that indicate the occurrence of a specific event.
[0003] However, existing research pays little attention to the multi-source nature of texts, and most of the work focuses on event detection in the context of single-source information, concentrating on the information contained in the single-source text itself. Therefore, there is a problem of incomplete event detection leading to inaccurate detection results. Summary of the Invention
[0004] The present invention was made to solve the above-mentioned problems, and its purpose is to provide an event detection method and apparatus that integrates multi-source information.
[0005] This invention provides an event detection method that integrates multi-source information. It is used to determine the event type of a specified event based on specified text containing a specified event and multiple existing texts. The method includes the following steps: Step S1, preprocessing the specified text and each existing text to obtain corresponding preprocessed specified text and preprocessed existing text; Step S2, inputting the preprocessed specified text and existing trigger word extraction templates into a candidate trigger word extraction model to obtain candidate trigger words corresponding to the preprocessed specified text; Step S3, performing text matching between the preprocessed specified text and each preprocessed existing text to determine if a matching text exists. If so, the matching text is then compared with the preprocessed specified text. The text matching preprocessing process uses existing text as the matching text and proceeds to step S4. If not, it uses candidate trigger words as trigger words and proceeds to step S7. In step S4, for each matching text, the matching text and the trigger word extraction template are input together into the candidate trigger word extraction model to obtain the candidate trigger words corresponding to the matching text. In step S5, a candidate trigger word prompt template is constructed based on the candidate trigger words corresponding to each matching text. In step S6, each matching text, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger words corresponding to the preprocessed specified text are input together into the trigger word extraction model to obtain the trigger words. In step S7, the trigger words are input into the trigger word classification model to obtain the event type.
[0006] The event detection method that integrates multi-source information provided by the present invention may also have the following features: in step S2, the preprocessed specified text and the trigger word extraction template are concatenated and the concatenation result is used as the input of the candidate trigger word extraction model; in step S4, the matching text and the trigger word extraction template are concatenated and the concatenation result is used as the input of the candidate trigger word extraction model; in step S6, all matching texts, candidate trigger word prompt templates and trigger word extraction templates containing candidate trigger words corresponding to the preprocessed specified text are concatenated and the concatenation result is used as the input of the trigger word extraction model.
[0007] The event detection method that integrates multi-source information provided by the present invention may also have the following feature: the candidate trigger word extraction model and the trigger word extraction model are both constructed based on existing generative pre-trained language models.
[0008] The event detection method that integrates multi-source information provided by the present invention may also have the following feature: wherein, in step S5, the candidate trigger word prompt template contains all candidate trigger words corresponding to each matching text.
[0009] The event detection method that integrates multi-source information provided by this invention may also have the following features: the trigger word classification model includes a BERT pre-trained language model and a multilayer perceptron. The BERT pre-trained language model generates the hidden layer representation vector of the sentence based on the trigger word, and the multilayer perceptron maps the hidden layer representation vector of the sentence to the label dimension to obtain the event type.
[0010] The event detection method that integrates multi-source information provided by the present invention may also have the following feature: wherein, in step S1, the preprocessing includes removing redundant non-text content from the text.
[0011] The event detection method integrating multi-source information provided by this invention may also have the following features: the process of training a candidate trigger word extraction model, a trigger word extraction model, and a trigger word classification model based on multiple existing training texts and their corresponding target candidate trigger words, target trigger words, and event type labels includes the following steps: Step T1, preprocessing all existing training texts and constructing a training dataset by combining the corresponding real trigger words and event type labels; Step T2, selecting existing training texts from the training dataset and inputting them into the candidate trigger word extraction model to generate predicted candidate trigger words, and fine-tuning the candidate trigger word extraction model by calculating the loss based on the predicted candidate trigger words and the corresponding target candidate trigger words; Step T3, repeating Step T2 until the first preset termination condition is reached, thus obtaining the trained candidate trigger word extraction model, and proceeding to Step T4; Step T4: Select existing training text from the training dataset, combine it with the trained candidate trigger word extraction model to construct the corresponding input data trigger word extraction model to generate predicted trigger words, and calculate the loss based on the predicted trigger words and the corresponding target trigger words to fine-tune the trigger word extraction model; Step T5: Repeat step T4 until the second preset termination condition is reached, then the trained trigger word extraction model is obtained, and proceed to step T6; Step T6: Select existing training text from the training dataset, combine it with the trained candidate trigger word extraction model and the trained trigger word extraction model to construct the corresponding input data trigger word classification model to generate predicted labels, and calculate the loss based on the predicted labels and the corresponding event type labels to fine-tune the trigger word classification model; Step T7: Repeat step T6 until the third preset termination condition is reached, then the trained trigger word classification model is obtained.
[0012] This invention also provides an event detection device that integrates multi-source information, used to obtain the event type of the specified event based on specified text containing the specified event and multiple existing texts. It features the following characteristics: a preprocessing module, used to preprocess the specified text and each existing text separately to obtain corresponding preprocessed specified text and preprocessed existing text; a candidate trigger word generation module, including a candidate trigger word extraction model and existing trigger word extraction templates, used to input the text and trigger word extraction templates together into the candidate trigger word extraction model to obtain candidate trigger words corresponding to the text, where the text includes the preprocessed specified text and matching text; and a matching judgment module, used to match the preprocessed specified text with each existing text. The system performs text matching on existing preprocessed text to determine if a matching text exists. If so, the existing preprocessed text that matches the specified preprocessed text is taken as the matching text; otherwise, the candidate trigger word is taken as the trigger word. A template construction module is used to construct candidate trigger word prompt templates based on the candidate trigger words corresponding to each matching text. A trigger word generation module, containing a trigger word extraction model, is used to input each matching text, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger words corresponding to the specified preprocessed text into the trigger word extraction model to obtain the trigger word. An event type generation module, containing a trigger word classification model, is used to input the trigger word into the trigger word classification model to obtain the event type.
[0013] The role and effect of invention
[0014] According to the event detection method and apparatus that integrates multi-source information of the present invention, on the one hand, matching text with the same specified event as the specified text is selected from existing text through text matching; on the other hand, trigger words are generated by combining the matching text and the specified text through trigger word extraction template, candidate trigger word prompt template, candidate trigger word extraction model, and trigger word extraction model, and the trigger words are input into the trigger word classification model to obtain the corresponding event type. Therefore, the event detection method and apparatus that integrates multi-source information of the present invention can generate accurate event detection results. Attached Figure Description
[0015] Figure 1 is a block diagram of an event detection device in an embodiment of the present invention;
[0016] Figure 2 is a flowchart illustrating the training of the candidate trigger word extraction model, the trigger word extraction model, and the trigger word classification model in an embodiment of the present invention.
[0017] Figure 3 is a flowchart illustrating the event detection method that integrates multi-source information in an embodiment of the present invention. Detailed Implementation
[0018] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the event detection method and apparatus for integrating multi-source information of the present invention.
[0019] This embodiment provides an event detection device that integrates multi-source information, hereinafter referred to as the event detection device, used to determine the event type of a specified event based on specified text containing a specified event and multiple existing texts. The specified text and existing texts corresponding to the event detection device are texts from the relevant domain selected according to actual needs.
[0020] Among texts from various fields, financial texts have several unique characteristics compared to other types of texts. Specifically, in today's internet age, financial texts often generate multiple different texts containing the same event during the dissemination process, i.e., multi-source information. For example, multiple news media outlets may produce multiple different reports on the same financial event, but the news events contained within them are consistent.
[0021] Therefore, in this embodiment, a financial event is used as the designated event, a news report containing the designated financial event is used as the designated text, and multiple other news reports are used as existing text to describe the event detection device of this embodiment.
[0022] Figure 1 is a block diagram of an event detection device in an embodiment of the present invention.
[0023] As shown in Figure 1, the event detection device 100 includes a preprocessing module 11, a candidate trigger word generation module 12, a matching judgment module 13, a template construction module 14, a trigger word generation module 15, an event type generation module 16, and a general control module 17 that controls the above modules.
[0024] The preprocessing module 11 is used to preprocess the specified text and each existing text separately to obtain the corresponding preprocessed specified text and preprocessed existing text.
[0025] The preprocessing includes removing redundant non-text content from the specified text and various existing texts. In this embodiment, the preprocessing also includes format conversion of the text to conform to the input format of the subsequent model.
[0026] The candidate trigger word generation module 12 includes a candidate trigger word extraction model and an existing trigger word extraction template. It is used to input the text and the trigger word extraction template into the candidate trigger word extraction model to obtain the candidate trigger words corresponding to the text. The text includes preprocessed specified text and matching text.
[0027] When the text is preprocessed specified text, the preprocessed specified text and the trigger word extraction template are concatenated, and the concatenated result is used as the input to the candidate trigger word extraction model. When the text is matched text, the matched text and the trigger word extraction template are concatenated, and the concatenated result is used as the input to the candidate trigger word extraction model.
[0028] In this embodiment, the trigger word extraction template is formatted as "event trigger word is". <trigger>.",in" <trigger>The "[SEP]" symbol is a special placeholder. In this embodiment, the concatenation result uses the "[SEP]" separator to separate the text and the trigger word extraction template. The output of the trigger word extraction template in this embodiment is "Event trigger word is". <trigger>.",in" <trigger>" is a candidate trigger word, and when multiple candidate trigger words are generated, they are connected by " and ".
[0029] The candidate trigger word extraction model is constructed based on an existing generative pre-trained language model. This generative pre-trained language model consists of two parts: an encoder and a decoder. The encoder is used to encode and model the input sequence, and the decoder is used to generate the target sequence based on the output of the encoder.
[0030] The matching judgment module 13 is used to perform text matching between the preprocessed specified text and each preprocessed existing text to determine whether there is a matching text. If so, the preprocessed existing text that matches the preprocessed specified text is taken as the matching text; otherwise, the candidate trigger word is taken as the trigger word.
[0031] In this embodiment, the matching judgment module 13 comprehensively evaluates whether two texts contain the same event through semantic matching or other feature matching methods. If so, the two texts are mutually matched, that is, there is a matching text. If not, there is no matching text.
[0032] Template building module 14 is used to build a candidate trigger word suggestion template based on the candidate trigger words corresponding to each matched text. In this embodiment, the candidate trigger word suggestion template is "the event trigger word may be..."<TriggerFrom Same Event> .",in"<TriggerFrom Same Event> " represents the corresponding candidate trigger word, and when there are multiple candidate trigger words corresponding to the matching text, the multiple candidate trigger words are connected by "and".
[0033] The candidate trigger word suggestion template contains all the candidate trigger words corresponding to each matched text.
[0034] The trigger word generation module 15 includes a trigger word extraction model, which is used to input each matching text, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger words corresponding to the preprocessed specified text into the trigger word extraction model to obtain the trigger words.
[0035] In this embodiment, all matching text, candidate trigger word suggestion templates, and trigger word extraction templates containing candidate trigger words corresponding to preprocessed specified text are concatenated, and the concatenation result is used as the input to the trigger word extraction model. In this embodiment, the matching text, candidate trigger word suggestion templates, and trigger word extraction templates are separated by the "[SEP]" interval identifier in the concatenation result.
[0036] The trigger word extraction model is constructed based on existing generative pre-trained language models.
[0037] The event type generation module 16 includes a trigger word classification model, which is used to input trigger words into the trigger word classification model to obtain event types.
[0038] The trigger word classification model includes a BERT pre-trained language model and a multilayer perceptron. The BERT pre-trained language model generates hidden layer representation vectors for the sentence based on the trigger words. The multilayer perceptron maps the hidden layer representation vectors of the sentence to the label dimension to obtain the event type.
[0039] In this embodiment, the candidate trigger word extraction model, the trigger word extraction model, and the trigger word classification model are trained sequentially in a specified order.
[0040] Figure 2 is a flowchart illustrating the training process of the candidate trigger word extraction model, the trigger word extraction model, and the trigger word classification model in an embodiment of the present invention.
[0041] As shown in Figure 2, the process of training a candidate trigger word extraction model, a trigger word extraction model, and a trigger word classification model based on multiple existing training texts and their corresponding target candidate trigger words, target trigger words, and event type labels includes the following steps:
[0042] Step T1: Preprocess all existing training texts and construct a training dataset by combining the corresponding real trigger words and event type labels.
[0043] Step T2: Select existing training text from the training dataset to input into the candidate trigger word extraction model to generate predicted candidate trigger words, and fine-tune the candidate trigger word extraction model by calculating the loss based on the predicted candidate trigger words and the corresponding target candidate trigger words.
[0044] Step T3: Repeat step T2 until the first preset termination condition is met. Then, the trained candidate trigger word extraction model is obtained, and the process proceeds to step T4.
[0045] Step T4: Select existing training texts from the training dataset, combine them with the trained candidate trigger word extraction model to construct the corresponding input data, input the trigger word extraction model to generate predicted trigger words, and calculate the loss based on the predicted trigger words and the corresponding target trigger words to fine-tune the trigger word extraction model.
[0046] Step T5: Repeat step T4 until the second preset termination condition is met, then the trained trigger word extraction model is obtained, and proceed to step T6.
[0047] Step T6 involves selecting existing training texts from the training dataset, combining the trained candidate trigger word extraction model with the trained trigger word extraction model to construct the corresponding input data trigger word classification model, generating predicted labels, and fine-tuning the trigger word classification model by calculating the loss based on the predicted labels and the corresponding event type labels. In this embodiment, the loss calculation for each model uses cross-entropy loss.
[0048] Step T7: Repeat step T6 until the third preset termination condition is reached, thus obtaining the trained trigger word classification model. In this embodiment, the first, second, and third preset termination conditions all refer to whether the number of iteration rounds has reached the corresponding maximum number of iteration rounds.
[0049] The main control module 17 stores the control program that controls the operation of each module.
[0050] The process of using the event detection device 100 to fuse multi-source information is described below with reference to the accompanying drawings.
[0051] Figure 3 is a flowchart illustrating the event detection method that integrates multi-source information in an embodiment of the present invention.
[0052] As shown in Figure 3, the event detection method that integrates multi-source information includes the following steps:
[0053] Step S1: The preprocessing module 11 is used to preprocess the specified text and each existing text to obtain the corresponding preprocessed specified text and preprocessed existing text.
[0054] Step S2: The candidate trigger word generation module 12 inputs the preprocessed specified text and the existing trigger word extraction template into the candidate trigger word extraction model to obtain the candidate trigger words corresponding to the preprocessed specified text.
[0055] Step S3: The matching judgment module 13 performs text matching judgment between the preprocessed specified text and each preprocessed existing text to determine whether there is a matching text. If so, the preprocessed existing text that matches the preprocessed specified text is taken as the matching text and proceeds to step S4. If not, the candidate trigger word is taken as the trigger word and proceeds to step S7.
[0056] Step S4: The candidate trigger word generation module 12 is used to input the matching text and the trigger word extraction template into the candidate trigger word extraction model to obtain the candidate trigger words corresponding to the matching text.
[0057] Step S5: The template building module 14 is used to build a candidate trigger word prompt template based on the candidate trigger words corresponding to each matched text.
[0058] Step S6: The trigger word generation module 15 inputs each matching text, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger words corresponding to the preprocessed specified text into the trigger word extraction model to obtain the trigger words.
[0059] Step S7: The event type generation module 16 is used to input the trigger word into the trigger word classification model to obtain the event type.
[0060] In this embodiment, existing data is used to perform performance tests on the Ours method (an event detection method that integrates multi-source information) compared with the existing Text2Event and DEGREE_ed methods on trigger word extraction and event classification tasks. The results are shown in the table below:
[0061]
[0062]
[0063] The first column in the table above represents the tasks, the second column represents the methods, and the third to fifth columns represent the precision, recall, and F1 scores of each method on each task, respectively. For example, the cell in the second row and third column shows that the Text2Event method has a precision of 72.97% on the trigger word extraction task. This demonstrates that the event detection method incorporating multi-source information has better trigger word extraction and event classification performance compared to existing methods.
[0064] The role and effect of the embodiments
[0065] According to the event detection method and apparatus involving multi-source information in this embodiment, on the one hand, matching text with the same specified event as the specified text is selected from existing text through text matching; on the other hand, trigger words are generated by combining the matching text and the specified text through trigger word extraction template, candidate trigger word prompt template, candidate trigger word extraction model and trigger word extraction model, and the trigger words are input into the trigger word classification model to obtain the corresponding event type. In summary, this method can generate accurate event detection results.
[0066] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.< / trigger> < / trigger> < / trigger> < / trigger>
Claims
1. An event detection method integrating multi-source information, used to obtain the event type of the specified event based on specified text containing the specified event and multiple existing texts, characterized in that, Includes the following steps: Step S1: Preprocess the specified text and each of the existing texts respectively to obtain the corresponding preprocessed specified text and preprocessed existing text; Step S2: Input the preprocessed specified text and the existing trigger word extraction template into the candidate trigger word extraction model to obtain the candidate trigger word corresponding to the preprocessed specified text; Step S3: Perform text matching between the preprocessed specified text and each of the existing preprocessed texts to determine if there is a matching text. If so, the existing preprocessed text that matches the preprocessed specified text is taken as the matching text and proceeds to Step S4. If not, the candidate trigger word is taken as the trigger word and proceeds to Step S7; Step S4: For each matching text, input the matching text and the trigger word extraction template into the candidate trigger word extraction model to obtain the candidate trigger word corresponding to the matching text; Step S5: Construct a candidate trigger word prompt template based on the candidate trigger words corresponding to each matching text; Step S6: Input each matching text, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger word corresponding to the preprocessed specified text into the trigger word extraction model to obtain the trigger word; Step S7: Input the trigger word into the trigger word classification model to obtain the event type.
2. The event detection method fusion of multi-source information according to claim 1, characterized in that: in, In step S2, the preprocessed specified text and the trigger word extraction template are concatenated, and the concatenation result is used as the input of the candidate trigger word extraction model. In step S4, the matching text and the trigger word extraction template are concatenated, and the concatenation result is used as the input of the candidate trigger word extraction model. In step S6, all the matching text, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger word corresponding to the preprocessed specified text are concatenated, and the concatenation result is used as the input of the trigger word extraction model.
3. The event detection method fusion of multi-source information according to claim 1, characterized in that: in, Both the candidate trigger word extraction model and the trigger word extraction model are constructed based on existing generative pre-trained language models.
4. The event detection method fusion of multi-source information according to claim 1, characterized in that: in, In step S5, the candidate trigger word suggestion template contains all the candidate trigger words corresponding to each of the matching texts.
5. The event detection method that integrates multi-source information according to claim 1, characterized in that: in, The trigger word classification model includes a BERT pre-trained language model and a multilayer perceptron. The BERT pre-trained language model generates a hidden layer representation vector of the sentence based on the trigger word, and the multilayer perceptron maps the hidden layer representation vector of the sentence to the label dimension to obtain the event type.
6. The event detection method fusion of multi-source information according to claim 1, characterized in that: in, In step S1, the preprocessing includes removing redundant non-text content from the text.
7. The event detection method fusion of multi-source information according to claim 1, characterized in that: in, The process of training the candidate trigger word extraction model, the trigger word extraction model, and the trigger word classification model based on multiple existing training texts and their corresponding target candidate trigger words, target trigger words, and event type labels includes the following steps: Step T1, preprocessing all the existing training texts and constructing a training dataset by combining the corresponding target candidate trigger words and event type labels; Step T2, selecting existing training texts from the training dataset and inputting them into the candidate trigger word extraction model to generate predicted candidate trigger words, and fine-tuning the candidate trigger word extraction model by calculating the loss based on the predicted candidate trigger words and the corresponding target candidate trigger words; Step T3, repeating Step T2 until a first preset termination condition is reached, thus obtaining the trained candidate trigger word extraction model, and proceeding to Step T4; Step T4, selecting existing training texts from the training dataset and inputting them into the candidate trigger word extraction model to generate predicted candidate trigger words, and calculating the loss based on the predicted candidate trigger words and the corresponding target candidate trigger words; Step T3, repeating Step T2 until a first preset termination condition is reached, thus obtaining the trained candidate trigger word extraction model, and proceeding to Step T4; Step T4, selecting existing training texts from the training dataset and inputting them into the candidate trigger word extraction model to generate predicted candidate trigger words, and then proceeding to Step T4. The text is combined with the trained candidate trigger word extraction model to construct corresponding input data, which is then input into the trigger word extraction model to generate predicted trigger words. The loss is calculated based on the predicted trigger words and the corresponding target trigger words to fine-tune the trigger word extraction model. Step T5: Step T4 is repeated until a second preset termination condition is met, resulting in a trained trigger word extraction model, and the process proceeds to step T6. Step T6: Existing training text is selected from the training dataset, and the corresponding input data is combined with the trained candidate trigger word extraction model and the trained trigger word extraction model to input into the trigger word classification model to generate predicted labels. The loss is calculated based on the predicted labels and the corresponding event type labels to fine-tune the trigger word classification model. Step T7: Step T6 is repeated until a third preset termination condition is met, resulting in a trained trigger word classification model.
8. An event detection device that integrates multi-source information, used to obtain the event type of the specified event based on specified text containing the specified event and multiple existing texts, characterized in that, include: The preprocessing module is used to preprocess the specified text and each of the existing texts respectively to obtain the corresponding preprocessed specified text and preprocessed existing text; The candidate trigger word generation module includes a candidate trigger word extraction model and an existing trigger word extraction template. It is used to input the text and the trigger word extraction template into the candidate trigger word extraction model to obtain the candidate trigger words corresponding to the text. The text includes the preprocessed specified text and the matching text. The matching judgment module is used to perform text matching judgment between the preprocessed specified text and each of the preprocessed existing texts to determine whether there is a matching text. If so, the preprocessed existing text that matches the preprocessed specified text is taken as the matching text; if not, the candidate trigger word is taken as the trigger word. The template construction module is used to construct a candidate trigger word prompt template based on the candidate trigger words corresponding to each of the matched texts; the trigger word generation module includes a trigger word extraction model, which is used to input each of the matched texts, the candidate trigger word prompt template, and the trigger word extraction template containing the candidate trigger words corresponding to the preprocessed specified text into the trigger word extraction model to obtain the trigger word; The event type generation module includes a trigger word classification model, which is used to input the trigger word into the trigger word classification model to obtain the event type.