A multi-document summary generation method for crisis help information

Abstract

The present invention discloses a multi-document summary generation method for crisis help information; the method is divided into: an extractive summary stage and a generative summary stage. In the extractive summary stage, the sentence graph structure is used to fuse contextual information in a deep neural network, and the subgraph structure is used to optimize the information extraction to effectively extract key information from numerous documents; in the generative summary stage, the powerful sequence generation ability of the BART model and the characteristics of the pointer generation network in accurately copying key information and avoiding duplicate content generation are used to further streamline and summarize the extracted information, and generate an accurate and compact summary of the crisis help information. The method of the present invention is suitable for processing and summarizing massive help information in crisis scenarios. The two-stage method of the present invention can quickly and accurately provide clear information summaries for rescue and emergency management, greatly improving the efficiency and effectiveness of crisis response.

Description

Technical Field

[0001] The present invention belongs to the technical field of text processing, and in particular relates to a multi-document summary generation method for crisis help information. Background Art

[0002] (1) Crisis information processing technology

[0003] During crises and disasters, social media has become a vital channel for information dissemination, carrying a wide range of information crucial for rescue efforts. Effectively parsing and managing this information is crucial for the success of rescue operations. To address this challenge, a range of research has focused on identifying, classifying, and processing crisis-related information from social media content.

[0004] Among them, one study aimed to identify and classify crisis-related content from social media tweets. By proposing an algorithm that combines traditional support vector machines, Gini index decision trees, and naive Bayes, it improved overall recognition performance. This algorithm was successfully applied to social media tweet data from the Indonesian earthquake, demonstrating high feasibility and accuracy in real-world scenarios. Another study extracted grammatical features from a dataset of online public opinion events to construct a grammatical feature library for events. By representing event text as syntactic vectors and using similarity recognition, this approach demonstrated high recognition accuracy for natural disasters, even with a small sample size. Furthermore, another study focused on identifying and extracting help-seeking messages from social media during flood events. This framework combined BERT-CNN technology with disaster knowledge and fine-tuned a pre-trained language model to successfully automatically extract help-seeking messages. Using a local rainstorm incident as an example, the model was able to automatically identify key help-seeking messages, such as trapped individuals and scarce supplies, with a small sample size.

[0005] Although many studies have made progress in the micro-extraction and identification of crisis information, how to better integrate information during a crisis to support decision-making and response is still in the exploratory stage. This is an important direction for future research, namely, to develop more efficient methods to comprehensively process and analyze the large amount of data generated in crisis situations in order to provide more accurate and timely information for rescue operations and disaster response.

[0006] (2) Pointer Network Technology

[0007] Pointer Networks (PNs) address the limitation encountered when the elements of the output sequence depend on the position of elements in the input sequence. In traditional models, the dimensionality of the output sequence is often limited by the size of the dictionary, making them ineffective for dealing with unknown words. In contrast, Pointer Networks incorporate an attention mechanism, enabling the model to learn the relative position of elements in the input sequence, allowing the output to directly point to a specific position in the input sequence.

[0008] The core components of a pointer network include an encoder and a decoder, which are typically constructed using a recurrent neural network. In this architecture, the encoder encodes the input sequence into a vector representation, while the decoder generates the current output based on the encoder's output and the previous state. The decoder uses a pointer mechanism to calculate and select the probability of a particular element in the input sequence as the output. This allows the model to directly output elements in the sequence based on the calculated probabilities, effectively solving the problem of unknown words. Summary of the Invention

[0009] In response to the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a multi-document summary generation method for crisis help information, which is suitable for processing and summarizing massive help information in crisis scenarios; through this two-stage method, the present invention can quickly and accurately provide clear information summaries for rescue and emergency management, greatly improving the efficiency and effectiveness of crisis response.

[0010] The technical solution of the present invention is specifically described as follows.

[0011] The present invention provides a multi-document summary generation method for crisis help information, which includes two stages: an extractive summary stage and a generative summary stage;

[0012] (1) Extractive summarization stage: extract key information from multiple documents through a crisis information extractive summarization model based on subgraph selection;

[0013] The subgraph selection-based crisis information extractive summarization model includes a graph-based multi-document encoder and a subgraph-based sentence selection module. The graph-based multi-document encoder utilizes the sentence graph structure to fuse contextual information in a deep neural network, and includes a text encoder, a graph encoder, and a subgraph encoder. The subgraph-based sentence selection module optimizes information extraction through the subgraph structure, and includes a graph pooling layer and a subgraph selection layer.

[0014] First, a document collection is input and passed through a text encoder to obtain the vector representation corresponding to each sentence. Then, a graph encoder converts the document collection into a relational graph using two graph construction methods: a similarity graph and a topic graph. The vector representation of each sentence constitutes a node in the relational graph, and the relationships between sentences form edges in the relational graph. Next, the graph encoder captures the relationships between sentences through this explicit sentence relational graph and integrates the relationships between sentences into the neural network through the graph structure. Subsequently, the subgraph encoder further models the subgraph structure based on the output of the graph encoder and uses the ROUGE score as the extraction criterion to construct a set of candidate summaries using a greedy algorithm. The graph pooling layer calculates the global pooling representation of the full graph and subgraphs. The subgraph selection layer selects the subgraph with the highest ROUGE score as the final summary.

[0015] (2) Generative summarization stage: The generative summarization model based on BART-PGN further summarizes and simplifies the extracted information to generate an accurate and compact summary of the crisis help information;

[0016] The generative summarization model is based on BART and improves the probability distribution mechanism of words generated by its autoregressive decoder by introducing a pointer generation network. The BART model adopts an encoder-decoder architecture.

[0017] First, a pointer generation mechanism is used so that the model can consider not only the possibility of generating new words through the vocabulary, but also the possibility of directly copying words from the input text; second, a coverage mechanism is used to accumulate the attention distribution in each time step and reduce the generation of repeated content by paying attention to the coverage vector.

[0018] In the present invention, the text encoder adopts the BERT pre-trained language model, and the graph encoder and subgraph encoder have the same structure, including a graph self-attention layer, a residual layer, a normalization layer and a feedforward layer.

[0019] In the present invention, the graph pooling layer uses a multi-head pooling mechanism, takes a set of sentence representations as input, and outputs a global pooled representation P.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] In the extractive summarization stage, the present invention designs an extraction method based on subgraph selection, which can effectively extract key information from numerous documents. It not only uses the sentence graph structure to fuse contextual information in the deep neural network, but also optimizes the information extraction through the subgraph structure, significantly improving the coherence and information richness of the summary.

[0022] In the generative summary stage, the present invention designs a model that combines the advantages of the BART model and the pointer generation network. By leveraging the powerful sequence generation capability of the BART model and the characteristics of the pointer generation network in accurately replicating key information and avoiding duplicate content generation, the extracted information can be further streamlined and summarized to generate an accurate and compact crisis help information summary. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a diagram of the extractive model structure of the present invention.

[0024] Figure 2 This is a diagram illustrating the structure of the encoder of the present invention.

[0025] Figure 3 Schematic diagram of the generative model structure of the present invention. DETAILED DESCRIPTION

[0026] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0027] The present invention proposes a multi-document summary generation method for crisis help information, which is divided into two stages. The specific steps are as follows:

[0028] (1) In response to the massive amount of crisis help-seeking texts, a crisis information extraction summarization model based on subgraph selection is used in the first stage of the processing process to extract key sentences from each document;

[0029] (2) Using a generative summarization model based on BART-PGN, the extracted key sentences are further streamlined and summarized to generate a highly generalized summary of the crisis help information. Ultimately, a comprehensive summary of massive crisis help information can be generated.

[0030] 1. Crisis Information Extractive Summarization Model Based on Subgraph Selection

[0031] The overall structure of the model is as follows Figure 1 As shown in Figure 2, the model is divided into two parts: graph-based multi-document encoder and subgraph-based sentence selection.

[0032] First, the model takes a document collection as input and, after passing it through the BERT model, obtains a vector representation corresponding to each sentence. Subsequently, the document collection is converted into a relational graph using two graph construction methods: a similarity graph and a topic graph. The vector representation of each sentence constitutes a node in the relational graph, and the relationships between sentences form edges in the graph. Next, the graph encoder captures the relationships between sentences using this explicit sentence relational graph and integrates these relationships into the neural network through the graph structure. Based on the output of the graph encoder, the model further models the subgraph structure using a subgraph encoder and calculates a global pooled representation of the full graph and subgraphs using a graph pooling layer. This process can distinguish the quality of different subgraph candidate summaries. Finally, the subgraph selection layer selects the subgraph with the highest score as the final summary.

[0033] 1. Graph-based Multi-Document Encoder

[0034] The graph-based multi-document encoder consists of three parts: text encoder, graph encoder and subgraph encoder.

[0035] (1) Text Encoder

[0036] The text encoder processes the document text and outputs corresponding sentence representations. In this method, the document text is encoded using the BERT pre-trained language model. Specifically, [CLS] and [SEP] tags are first inserted at the beginning and end of each sentence, and then fed into the BERT model to obtain sentence representations.

[0037] (2) Image Encoder

[0038] The graph encoder updates the sentence representation by modeling the relationship graph between sentences. The graph encoder structure is as follows Figure 2 As shown in the figure, the input first passes through the graph self-attention layer, then performs residual addition and layer normalization on the input. The output is then passed to the feedforward layer for similar residual and layer normalization processing. The output of the graph encoder is a sentence vector that incorporates the graph relationship between sentences.

[0039] In the graph encoder, we first need to obtain the sentence relationship graph. In the sentence relationship graph, sentences are regarded as nodes of the graph, and the relationships between sentences are regarded as edges. Therefore, we use the graph representation matrix G to represent the graph structure information. Specifically, the element G[i][j] in the matrix represents the sentence s i With sentence s j The relationship weight E ij ,Relationship weights use different standard methods in different graph types (similarity graph and topic graph).

[0040] Then, the graph-aware self-attention mechanism is used to enable the model to more effectively capture and understand the relationship between sentences and other related sentences. The graph-aware self-attention mechanism extends the self-attention mechanism to consider the relationship between sentences in the explicit sentence relationship graph. For sentence s i At the output of the l-1th graph encoding layer, u i For sentence s i The context representation, e ij Represents sentence s i and s j The original self-attention weight between ij represents the adjusted weight. Then its calculation method is as follows:

[0041]

[0042]

[0043] α ij =softmax(e ij +R ij )

[0044] The key to the graph-aware self-attention mechanism lies in the bias R of the relationship between sentences. ij , that is, this step, integrates the sentence relationship graph into the neural network. ij The calculation method is as follows:

[0045]

[0046] Next, the result of the operation will be passed to a feedforward neural network and then subjected to residual connection and layer normalization. This feedforward neural network contains two linear layers and ReLU activation function. Through this process, the vector representation x of each sentence will be obtained. l .

[0047] (3) Sub-image encoder

[0048] The subgraph encoder is used to model each subgraph, and its architecture is exactly the same as that of the graph encoder. The input of the subgraph encoder is the output of the graph encoder; the output of the subgraph encoder is a subgraph vector consisting of a set of sentences.

[0049] There are two challenges in training a subgraph encoder: (1) The gold abstracts for most datasets are written by humans and need to be converted into sentence tokens for extractive summaries. (2) There are hundreds of sentences in a document collection, and thousands of subgraphs to choose from, so the efficiency and performance of subgraph construction must be guaranteed. To address these two issues, this paper uses the ROUGE score as the extraction criterion and a greedy algorithm to construct a set of candidate abstracts.

[0050] Specifically, we first select several sentences with the highest ROUGE scores as initial candidate nodes, which serve as the initial nodes of multiple candidate subgraphs. We then gradually add sentences to the summary set, maximizing the ROUGE score of the currently selected sentence set relative to the golden summary, until adding any remaining sentences no longer improves the ROUGE score. Ultimately, these sentence sets correspond to multiple constructed subgraphs. The algorithm for this process is shown in Table 1:

[0051] Table 1

[0052]

[0053]

[0054] 2. Subgraph-based statement selection

[0055] Subgraph-based sentence selection consists of two parts: graph pooling layer and subgraph selection layer.

[0056] (1) Graph pooling layer

[0057] The main goal of the graph pooling layer is to capture the overall semantic information of the entire sentence relationship graph and each subgraph. The graph pooling layer uses a multi-head pooling mechanism, takes a set of sentence representations as input, and outputs a global pooled representation P.

[0058] For each independent attention head z∈{1,…,n head}, first calculate the attention score and value vector for the vector representation of the sentence, and calculate the attention distribution of each sentence based on the attention score. Then perform weighted summation on each attention head, and obtain the vector representation corresponding to each attention head through linear transformation and layer normalization. z Finally, the vectors of all attention heads are concatenated and linearly transformed to obtain the final global pooling representation P. Therefore, the global pooling representation is calculated as follows:

[0059]

[0060] P=W d [head1||…||head z ]

[0061] (2) Sub-image selection layer

[0062] The subgraph selection layer scores each subgraph to select the best subgraph as the final summary. Two losses are used as the loss function of the entire model. Among them, cosine similarity is used to calculate the similarity between the golden summary C* and the global pooling representation P as the loss function of the summary task, and binary cross entropy loss is used to learn more accurate sentence and summary representations. Assume Indicates whether the sentence should be considered as a summary sentence, then the overall loss function is calculated as follows:

[0063]

[0064] 2. Generative Summarization Model Based on BART-PGN

[0065] The BART model uses an encoder-decoder architecture. Based on BART, the generative summary model improves the probability distribution mechanism of words generated by its autoregressive decoder by introducing a pointer generation network. First, the pointer generation mechanism is used so that the model can consider not only the possibility of generating new words through the vocabulary, but also the possibility of directly copying words from the input text. Secondly, the coverage mechanism is used to accumulate the attention distribution in each time step, and reduce the generation of repeated content by paying attention to the coverage vector. The overall structure of the model is as follows Figure 3 shown.

[0066] a) Pointer generation mechanism

[0067] First, at each moment, the model is divided into generation mode and replication mode. In generation mode, the model needs to generate words from the vocabulary. This process allows the model to introduce new words when generating text. The hidden state and context vector output by the decoder are concatenated, and after two linear transformations and a softmax layer, the probability distribution P of the vocabulary at that moment is obtained. vocab In copy mode, the model can directly select and copy words from the input sequence. The model will be based on the current moment's attention distribution. To determine the probability P of copying word w in the input sequence copy (w), is calculated as follows:

[0068]

[0069] In order to choose between the generation mode and the replication mode, a generation probability P is introduced. gen As a soft switch, it determines the current decoding method, whether to generate new words or copy words directly from the input sequence. The soft switch is calculated by the context vector, the decoder hidden state, and the decoder input after passing it through the Sigmoid function.

[0070] Finally, at each moment, the model performs a weighted calculation on the probability distribution obtained in the generation mode and the probability distribution obtained in the replication mode to obtain the probability distribution P(w) of the decoded word. The calculation method is as follows:

[0071] P(w)=P gen P vocab (w)+(1-P gen )Pcopy (w)

[0072] b) Coverage mechanism

[0073] The core of the coverage mechanism is to adjust the attention mechanism to prevent the probability of attention distribution of a certain word in the original text from being too high, resulting in too large a weight. It can effectively reduce or avoid repeated concentration of attention on the same position, thereby significantly reducing the repeated content in the generated summary.

[0074] Specifically, through the coverage vector c t Attention distribution for generated words Penalize, the coverage vector value is the sum of the attention distribution at time t-1 and before. Assume that W h 、W o 、W c represents the linear transformation matrix, v T represents the weight vector, represents the coverage vector, b attn Represents the bias matrix, the original attention distribution is output by the encoder h i and decoder output o t Calculated, the adjusted coverage vector and attention distribution are calculated as follows:

[0075]

[0076]

[0077] Implementation examples:

[0078] This example involves developing a crisis help platform system, designed to provide users with a platform for posting and accessing crisis-related help requests and supporting large-scale data processing. The system is programmed in JavaScript, Golang, and Python, and the system and the model proposed in this article are deployed on a Linux server. The system can periodically generate summaries of help requests related to a specific crisis event. The specific steps are as follows: First, user-posted crisis help requests are preprocessed to improve data quality and prepare for summary generation. The preprocessed text data is then input into the summary model proposed in this article to generate a coherent and informative summary. Finally, the resulting crisis information summary is presented to the public through a user interface. The inventors conducted experiments, and the results showed that the summary results of the present invention were significantly superior to those of multiple traditional summary models on ROUGE-1, ROUGE-2, and ROUGE-L.

Claims

1. A multi-document summary generation method for crisis assistance information, characterized by: The method consists of two stages: extractive summarization stage and generative summarization stage; (1) Extractive summarization stage: extract key information from multiple documents through a crisis information extractive summarization model based on subgraph selection; The subgraph selection-based crisis information extractive summarization model includes a graph-based multi-document encoder and a subgraph-based sentence selection module. The graph-based multi-document encoder utilizes the sentence graph structure to fuse contextual information in a deep neural network, which includes a text encoder, a graph encoder, and a subgraph encoder. The subgraph-based sentence selection module optimizes information extraction through the subgraph structure, which includes a graph pooling layer and a subgraph selection layer; First, a document collection is input and passed through a text encoder to obtain the vector representation corresponding to each sentence. Then, a graph encoder converts the document collection into a relational graph using two graph construction methods: a similarity graph and a topic graph. The vector representation of each sentence constitutes a node in the relational graph, and the relationships between sentences form edges in the relational graph. Next, the graph encoder captures the relationships between sentences through this explicit sentence relational graph and integrates the relationships between sentences into the neural network through the graph structure. Subsequently, the subgraph encoder further models the subgraph structure based on the output of the graph encoder and uses the ROUGE score as the extraction criterion to construct a set of candidate summaries using a greedy algorithm. The graph pooling layer calculates the global pooling representation of the full graph and subgraphs. The subgraph selection layer selects the subgraph with the highest ROUGE score as the final summary. (2) Generative summarization stage: The generative summarization model based on BART-PGN further summarizes and simplifies the extracted information to generate an accurate and compact summary of the crisis help information; The generative summarization model is based on BART and improves the probability distribution mechanism of words generated by its autoregressive decoder by introducing a pointer generation network. The BART model adopts an encoder-decoder architecture. First, a pointer generation mechanism is used so that the model can consider not only the possibility of generating new words through the vocabulary, but also the possibility of directly copying words from the input text; second, a coverage mechanism is used to accumulate the attention distribution in each time step and reduce the generation of repeated content by paying attention to the coverage vector.

2. The multi-document summary generation method according to claim 1, characterized in that: The text encoder uses the BERT pre-trained language model. The graph encoder and subgraph encoder have the same structure, including a graph self-attention layer, a residual layer, a normalization layer, and a feedforward layer.

3. The multi-document summary generation method according to claim 1, characterized in that: In the graph encoder, we first construct a sentence relationship graph; In the sentence relationship graph, sentences are regarded as nodes of the graph, and the relationships between sentences are regarded as edges. Therefore, the graph representation matrix G is used to represent the graph structure information; The element G[i][j] in the matrix represents the sentence s i With sentence s j The relationship weight E ij ; Then, a graph-aware self-attention mechanism is used to consider the relationships between sentences in the explicit sentence relation graph; assuming For sentence s i At the output of the l-1th graph encoding layer, u i For sentence s i The context representation, e ij Represents sentence s i and s j The original self-attention weight between Q 、W K 、W V They are the Query, Key, and Value weight matrices in the self-attention mechanism, d head is the attention head, α ij represents the adjusted weight, α ij The calculation method is as follows: a ij =softmax(e ij +R ij ) where R ij The bias that represents the relationship between sentences, R ij The calculation method is as follows: The graph-aware self-attention mechanism integrates sentence relationship graphs into deep neural networks by extending the self-attention mechanism; Next, the adjusted weight α of the sentence ij Passed into the feedforward neural network and then subjected to residual connection and layer normalization to obtain the vector representation x of each sentence l .

4. The multi-document summary generation method according to claim 1, characterized in that: In the subgraph encoder, the method of using the greedy algorithm to construct the candidate summary set is as follows: First, several sentences with the highest ROUGE scores are selected as initial candidate nodes, which are then used as initial nodes for multiple candidate subgraphs. Then, gradually add sentences to the summary set so that the ROUGE score of the currently selected sentence set is maximized relative to the golden summary C*, until adding any remaining sentences does not improve the ROUGE score; Finally, these sentence sets correspond to multiple constructed subgraphs.

5. The multi-document summary generation method according to claim 1, characterized in that: The graph pooling layer uses a multi-head pooling mechanism, takes a set of sentence representations as input, and outputs a global pooled representation P.

6. The multi-document summary generation method according to any one of claims 1 to 5, characterized in that: Two losses are used as the overall loss function of the crisis information extractive summary model based on subgraph selection; among them, cosine similarity is used to calculate the similarity between the golden summary C* and the global pooling representation P as the loss function of the summary task, and binary cross entropy loss is used to learn more accurate sentence and summary representations; assuming y i represents the true label of the sentence, Indicates whether the sentence should be considered as a summary sentence, then the overall loss function is calculated as follows: .

7. The multi-document summary generation method according to claim 1, characterized in that: The generative summarization model based on BART-PGN is divided into generation mode and copy mode. When generating text, the generation mode generates new words from the vocabulary, and the copy mode selects and copies words from the input sequence. Improvements are made to the probability distribution mechanism of words generated by its autoregressive decoder, including: Using the pointer generation mechanism, the probability distribution of decoded words at each moment is obtained by considering the probability of generating new words from the vocabulary and the probability of directly copying words from the input text at each moment. Using the coverage mechanism, we accumulate the attention distribution at each time step and reduce the generation of duplicate content by focusing on the coverage vector.

8. The multi-document summary generation method according to claim 7, characterized in that: In the generative summary stage, at the current moment, the hidden state and context vector output by the decoder are concatenated, and after two linear transformations and a softmax layer, the probability distribution P of generating new words through the vocabulary at this moment is obtained. vocab ; According to the current attention distribution Determine the probability P of copying word w in the input sequence copy (w), is calculated as follows: In order to choose between the generation mode and the replication mode, a generation probability P is introduced. gen As a soft switch, it is used to determine which decoding method is currently used; The soft switch is calculated by passing the context vector, the decoder hidden state, and the decoder input through the Sigmoid function; Finally, at each moment, the model performs a weighted calculation on the probability distribution obtained in the generation mode and the probability distribution obtained in the replication mode to obtain the probability distribution P(w) of the decoded word; the calculation method is as follows: P(w)=P gen P vocab (w)+(1-P gen )P copy (w)。 9. The multi-document summary generation method according to claim 7, characterized in that: In the generative summary stage, the coverage vector c t Attention distribution for generated words Penalty is performed, and the value of the coverage vector is the sum of the attention distribution at time t-1 and before.

Images