Text summary analysis method, device, electronic device and storage medium

Abstract

The present invention provides a text summary analysis method, comprising: determining a self-reference redundancy parameter corresponding to the summary text; determining pseudo-reference information corresponding to the target text; determining a center weighting parameter that matches the target text based on the pseudo-reference information corresponding to the target text; determining a correlation parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighting parameter that matches the target text, thereby determining an analysis result of the summary text. The present invention also provides a text summary analysis device, an electronic device, and a storage medium. The present invention can reduce the impact of noise information in the pseudo-reference information on the analysis result of the summary text by using the center weighting parameter that matches the target text, and can more accurately and automatically process summary texts without manually annotated references by using the self-reference redundancy parameter corresponding to the summary text, thereby improving the efficiency of summary text evaluation.

Description

Technical Field

[0001] The present invention relates to information processing technology, and in particular to a text summary analysis method, device, electronic device and storage medium. Background Art

[0002] The summary of text information can be formed by extracting key content from text information through short text descriptions, such as news, medical records, papers, Weibo, chats, etc., thus saving users' time browsing text. Generating text summaries through artificial intelligence is an efficient solution. To this end, artificial intelligence technology (AI) provides an appropriate text processing process operation mechanism to support the above application solutions. Among them, artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making. In the field of text processing, digital computers or machines controlled by digital computers can be used to realize the recognition of text information.

[0003] Different machine learning technologies can generate different summaries of the same text, and their quality can also vary. Therefore, it's necessary to analyze summaries scientifically and accurately to determine their quality. Summary of the Invention

[0004] In view of this, the embodiments of the present invention provide a text summary analysis method, device, electronic device, and storage medium. The technical solution of the embodiments of the present invention is implemented as follows:

[0005] The embodiment of the present invention adopts a text summary analysis method, including:

[0006] Obtain target text and summary text in a text summary analysis environment;

[0007] Processing the summary text through a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text;

[0008] Processing the target text through a text information processing model to determine pseudo reference information corresponding to the target text;

[0009] Determining a center weighting parameter that matches the target text based on the pseudo reference information corresponding to the target text;

[0010] Determining a relevance parameter between the target text and the summary text based on a self-reference redundancy parameter corresponding to the summary text and a center weighted parameter matching the target text;

[0011] An analysis result of the summary text is determined based on the relevance parameter between the target text and the summary text.

[0012] An embodiment of the present invention further provides a text summary analysis device, comprising:

[0013] An information transmission module is used to obtain the target text and summary text in the text summary analysis environment;

[0014] An information processing module, configured to process the summary text using a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text;

[0015] The information processing module is used to process the target text through a text information processing model to determine pseudo reference information corresponding to the target text;

[0016] The information processing module is used to determine a center weighting parameter matching the target text based on the pseudo reference information corresponding to the target text;

[0017] The information processing module is configured to determine a relevance parameter between the target text and the summary text based on a self-reference redundancy parameter corresponding to the summary text and a center weighting parameter matching the target text;

[0018] The information processing module is used to determine the analysis result of the summary text based on the relevance parameter between the target text and the summary text.

[0019] In the above scheme,

[0020] The information processing module is used to process the summary text through the encoding network of the text information processing model and extract a summary feature vector that matches the summary text;

[0021] The information processing module is configured to determine a first word-level latent vector corresponding to the summary text based on the summary feature vector;

[0022] The information processing module is configured to perform maximum pooling processing on the first word-level latent vector to form a first sentence-level latent vector;

[0023] The information processing module is configured to concatenate the first word-level latent vector and the first sentence-level latent vector to form a first concatenated vector;

[0024] The information processing module is configured to determine, using the first concatenated vector, similarities between different words in the first word-level latent vector and similarities between different sentences in the first sentence-level latent vector;

[0025] The information processing module is used to determine the self-reference redundancy parameter corresponding to the summary text according to the similarity between different words and the similarity between different sentences.

[0026] In the above scheme,

[0027] The information processing module is configured to determine a second word-level latent vector corresponding to the target text based on the summary feature vector;

[0028] The information processing module is configured to perform pooling processing on the second word-level latent vector to form a second sentence-level latent vector;

[0029] The information processing module is configured to determine a centrality score corresponding to each sentence in the target text based on the second sentence-level latent vector;

[0030] The information processing module is used to sort the centrality scores corresponding to each sentence in the target text, and determine the pseudo-reference information corresponding to the target text according to the sorting results of the centrality scores.

[0031] In the above scheme,

[0032] The information processing module is configured to determine the number of words in the pseudo-reference information according to the text summary analysis environment;

[0033] The information processing module is configured to obtain a corresponding number of sentences from the ranking results of the centrality scores based on the number of words in the pseudo reference information as pseudo reference information corresponding to the target text.

[0034] In the above scheme,

[0035] The information processing module is configured to concatenate the second word-level latent vector and the second sentence-level latent vector to form a second concatenated vector;

[0036] The information processing module is used to determine the weight parameters corresponding to each sentence and word in the pseudo reference information corresponding to the target text;

[0037] The information processing module is used to determine a center weighting parameter that matches the target text based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information.

[0038] In the above scheme,

[0039] The information processing module is used to determine the centrality score corresponding to each sentence in the target text;

[0040] The information processing module is used to normalize the centrality score corresponding to each sentence in the target text to form a sentence-level centrality score;

[0041] The information processing module is used to determine the sentence to which each word in the pseudo reference information belongs;

[0042] The information processing module is used to determine a word-level centrality score based on the centrality score of the sentence to which each word in the pseudo-reference information belongs;

[0043] The information processing module is used to determine the weight parameters corresponding to each sentence and word in the pseudo-reference information according to the sentence-level centrality score and the word-level centrality score.

[0044] In the above scheme,

[0045] The information processing module is configured to determine an accuracy parameter corresponding to the target text using the first splicing vector based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information;

[0046] The information processing module is configured to determine a recall parameter corresponding to the target text using the first splicing vector by using the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information;

[0047] The information processing module is used to determine a center weighting parameter that matches the target text according to the accuracy parameter and the recall parameter.

[0048] In the above scheme,

[0049] The information processing module is configured to, when the weight of the recall parameter corresponding to the target text in the text summary analysis environment is less than or equal to the weight of the accuracy parameter, use the center weighted parameter that matches the target text as the relevance parameter between the target text and the summary text.

[0050] In the above scheme,

[0051] The information processing module is configured to, when the weight of the recall parameter corresponding to the target text in the text summary analysis environment is greater than the weight of the accuracy parameter,

[0052] The information processing module is used to determine the ratio of the weight of the recall rate parameter to the weight of the precision rate parameter;

[0053] The information processing module is used to determine the relevance parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text, by using the ratio of the weight of the recall rate parameter to the weight of the precision parameter.

[0054] In the above scheme,

[0055] The information processing module is used to trigger the corresponding word segmentation library according to the text parameter information carried by the target text;

[0056] The information processing module is used to perform word segmentation processing on the target text through the triggered word segmentation library dictionary to form different word-level feature vectors;

[0057] The information processing module is used to perform noise removal on the different word-level feature vectors to form a set of word-level feature vectors corresponding to the target text.

[0058] In the above scheme,

[0059] The information processing module is used to determine a dynamic noise threshold that matches the usage environment of the text information processing model;

[0060] The information processing module is configured to perform noise removal processing on the different word-level feature vectors according to the dynamic noise threshold, and trigger a dynamic word segmentation strategy that matches the dynamic noise threshold;

[0061] The information processing module is used to perform word segmentation processing on the target text according to a dynamic word segmentation strategy that matches the dynamic noise threshold, so as to form a set of dynamic word-level feature vectors corresponding to the target text.

[0062] In the above scheme,

[0063] The information processing module is used to determine a fixed noise threshold corresponding to the use environment of the text information processing model;

[0064] The information processing module is configured to perform noise removal processing on the different word-level feature vectors according to the fixed noise threshold, and trigger a fixed word segmentation strategy that matches the fixed noise threshold;

[0065] The information processing module is used to perform word segmentation processing on the target text according to a fixed word segmentation strategy that matches the fixed noise threshold, and a fixed word-level feature vector set corresponding to the target text.

[0066] In the above scheme,

[0067] The information processing module is used to perform word segmentation processing on the target text to form a word segmentation processing result;

[0068] The information processing module is configured to perform stop word removal processing on the target text in response to the word segmentation processing result to form text keywords that match the target text;

[0069] The information processing module is used to determine the part-of-speech tagging results that match the target text based on the text keywords that match the target text, and form a part-of-speech feature vector set corresponding to the target text.

[0070] An embodiment of the present invention further provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium to implement the aforementioned text summary analysis method.

[0071] An embodiment of the present invention further provides an electronic device, comprising:

[0072] a memory for storing executable instructions;

[0073] The processor is configured to implement the aforementioned text summary analysis method when executing the executable instructions stored in the memory.

[0074] An embodiment of the present invention further provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implement the aforementioned text summary analysis method.

[0075] The embodiments of the present invention have the following beneficial effects:

[0076] The method obtains a target text and a summary text in a text summary analysis environment; processes the summary text through a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text; processes the target text through a text information processing model to determine pseudo-reference information corresponding to the target text; determines a center weighting parameter matching the target text based on the pseudo-reference information corresponding to the target text; determines a relevance parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighting parameter matching the target text; and determines an analysis result of the summary text based on the relevance parameter between the target text and the summary text. Thus, not only is automated evaluation of the summary text achieved, but also, through the center weighting parameter matching the target text, the influence of noise information in the pseudo-reference information on the analysis result of the summary text is reduced; and through the self-reference redundancy parameter corresponding to the summary text, summary texts without manually annotated references are more accurately automatically processed, thereby improving the efficiency of summary text evaluation and enhancing the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A schematic diagram of a usage scenario of the text summary analysis method provided by an embodiment of the present invention;

[0078] Figure 2 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention;

[0079] Figure 3 An optional flowchart of the text summary analysis method provided by an embodiment of the present invention;

[0080] Figure 4 This is a schematic diagram of an optional data processing flow in the summary text evaluation process according to an embodiment of the present invention;

[0081] Figure 5 This is an optional structural diagram of a text information processing model in an embodiment of the present invention;

[0082] Figure 6 An optional flowchart of text summary analysis provided by an embodiment of the present invention;

[0083] Figure 7 This is a schematic diagram of an optional data processing flow in the summary text evaluation process according to an embodiment of the present invention;

[0084] Figure 8 It is a schematic diagram of processing different medical record text information;

[0085] Figure 9 The present invention provides an optional flowchart of a text summary analysis method. DETAILED DESCRIPTION

[0086] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0087] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0088] The relevant data collection and processing in the embodiments of this application should be strictly in accordance with the requirements of relevant laws and regulations when applied in examples, and the informed consent or separate consent of the personal information subject should be obtained. Subsequent data use and processing should be carried out within the scope of authorization of laws and regulations and the personal information subject.

[0089] Before further explaining the embodiments of the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0090] 1) In response, it is used to indicate the conditions or states on which the executed operations depend. When the dependent conditions or states are met, one or more operations executed can be real-time or have a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations executed are executed.

[0091] 2) Word segmentation library: also known as word segmentation library, refers to a specific word segmentation method. Different word segmentation libraries have corresponding word dictionaries, and can perform word segmentation on the corresponding text information based on their corresponding word dictionaries.

[0092] 3) Consistency: This means that the data accessed from different servers is always unique.

[0093] 4) Neural Network (NN): Artificial Neural Network (ANN), also known as neural network or quasi-neural network, is a mathematical model or computational model that imitates the structure and function of biological neural networks (the central nervous system of animals, especially the brain) in the fields of machine learning and cognitive science. It is used to estimate or approximate functions.

[0094] 5) Encoder-Decoder Architecture: This is a common network structure used in machine translation. It consists of two parts: an encoder and a decoder. The encoder converts the input text into a series of context vectors that represent the characteristics of the input text. The decoder receives the encoder's output as input and outputs the corresponding text sequence in another language.

[0095] 6) Mini Programs are programs developed using a front-end language (e.g., JavaScript) that implement services within a Hypertext Markup Language (HTML) page. These programs are downloaded by a client (e.g., a browser or any client with an embedded browser core) via a network (e.g., the Internet) and interpreted and executed within the client's browser environment, eliminating the need for client-side installation. For example, by waking up a mini program on a terminal through voice commands, a social network client can download and run mini programs for various services, such as ticket purchases, task processing and creation, and data presentation.

[0096] 7) BERT: Bidirectional Encoder Representations from Transformers, a language model training method that leverages massive amounts of text. This method is widely used in various natural language processing tasks, such as text classification, text matching, and machine reading comprehension.

[0097] 8) Summary: The summary text generated by the automatic text summarization system through different machine learning processes.

[0098] 9) Centrality-weighted relevance: A centrality-weighted relevance parameter used to characterize the relevance between text information and summary text by pseudo-referencing the centrality scores of each word and sentence, and determined by the ratio of precision to recall. Recall / precision: The precision and recall rates in the relevance score, where precision indicates the accuracy of the summary text and recall indicates the amount of information used in the summary text. A higher recall rate indicates more accurately extracted summary information. Self-referenced redundancy score: A self-referenced redundancy score parameter used to characterize the amount of redundant information in the summary text. The more redundant information, the larger the self-referenced redundancy score.

[0099] 10) Hidden representation: A continuous vector that represents the semantic information of the word or sentence it represents. Sentence-level / word-level: sentence-level and word-level.

[0100] 11) Pseudo reference: A text consisting of important sentences of text information, which serves as a reference for the summary to be evaluated, helping to reflect how much important information in the text information is contained in the summary.

[0101] 12) Centrality-based sentence selection: Centrality-based sentence selection, a method of selecting textual information sentences as pseudo-references.

[0102] 13) Relevance score: The relevance score represents how much important information the summary contains in the text and is one of the core indicators of summary quality.

[0103] 14) Averaged relevance score: average relevance score. When there are multiple text messages, the final relevance score of the summary is the average of the relevance scores of the summary and each text message.

[0104] Figure 1 Schematic diagram of the use scenario of the text summary analysis method provided by the embodiment of the present invention, see Figure 1, the terminal (including terminal 10-1 and terminal 10-2) is provided with a corresponding client that can perform different functions. Among them, the client is that the terminal (including terminal 10-1 and terminal 10-2) obtains different text information (including but not limited to: news articles, professional papers, public account articles, novels, medical records) from the corresponding server 200 through the network 300 for browsing. The terminal is connected to the server 200 through the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two, and a wireless link is used to realize data transmission. Among them, the types of text information obtained by the terminal (including terminal 10-1 and terminal 10-2) from the corresponding server 200 through the network 300 are not the same. For example: the terminal (including terminal 10-1 and terminal 10-2) can obtain any type of text information from the corresponding server 200 through the network 300, or obtain browsing summary information from the corresponding server 200 through the network 300 and then select the required text content for browsing. The server 200 can store different summary texts for the same target text generated by different machine learning processes through different text processing neural networks. The text summary analysis method provided by the present application can reduce the impact of noise information in pseudo-reference information on the analysis results of the summary text by using the center weighting parameter matching the target text. The self-reference redundancy parameter corresponding to the summary text can be used to more accurately automatically process summary texts without manually annotated references, thereby improving the efficiency of summary text evaluation and allowing readers to ultimately obtain summary texts that match the target text, thereby improving the processing efficiency of text information. In some embodiments of the present invention, the different types of text information stored in the server 200 can be written in different software code environments using programming languages ​​in different text summary analysis environments, and code objects can be different types of code entities. For example, in C language software code, a code object can be a function. In JAVA language software code, a code object can be a class, and in iOS OC language, it can be a section of object code. In C++ language software code, a code object can be a class or a function to execute text processing instructions from different terminals. In this application, the source of the text information to be processed by the text information processing model is no longer distinguished.

[0105] In the process of the server 200 sending different types of text information to the terminal (terminal 10-1 and / or terminal 10-2) through the network 300, the server 200 needs to determine the text information selected by the user for detection. Therefore, as an example, the server 200 is used to obtain the target text and the summary text in the text summary analysis environment; process the summary text through the text information processing model to determine the self-reference redundancy parameter corresponding to the summary text; process the target text through the text information processing model to determine the pseudo-reference information corresponding to the target text; based on the pseudo-reference information corresponding to the target text, determine the center weighting parameter matching the target text; based on the self-reference redundancy parameter corresponding to the summary text and the center weighting parameter matching the target text, determine the relevance parameter between the target text and the summary text; based on the relevance parameter between the target text and the summary text, determine the analysis result of the summary text, and finally present the matching summary text in the display interface to facilitate the user to perform different subsequent operations.

[0106] The structure of the electronic device of the embodiment of the present invention is described in detail below. The electronic device can be implemented in various forms, such as a dedicated terminal with a text processing function, or an electronic device with a text processing function, such as the aforementioned Figure 1 Server 200 in. Figure 2 The schematic diagram of the structure of the electronic device provided in the embodiment of the present invention can be understood as follows: Figure 2 Only exemplary structures of electronic devices are shown, not all structures, and may be implemented as needed. Figure 2 Partial or complete structure shown.

[0107] The electronic device provided in the embodiment of the present invention includes: at least one processor 201, a memory 202, a user interface 203 and at least one network interface 204. The various components in the electronic device 20 are coupled together via a bus system 205. It can be understood that the bus system 205 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 205 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 205 is not described in detail. Figure 2 Various buses are labeled as bus system 205 .

[0108] The user interface 203 may include a display, a keyboard, a mouse, a trackball, a click wheel, keys, buttons, a touch pad or a touch screen.

[0109] It is understood that the memory 202 can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The memory 202 in the embodiment of the present invention can store data to support the operation of the terminal (such as 10-1). Examples of such data include: any computer program used to operate on the terminal (such as 10-1), such as an operating system and an application program. Among them, the operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application program can include various application programs.

[0110] In some embodiments, the text summary analysis device provided by the embodiments of the present invention can be implemented using a combination of software and hardware. As an example, the text summary analysis device provided by the embodiments of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the text summary analysis method provided by the embodiments of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0111] As an example of a text summary analysis device provided by an embodiment of the present invention being implemented by combining software and hardware, the text summary analysis device provided by an embodiment of the present invention can be directly embodied as a combination of software modules executed by the processor 201. The software modules can be located in a storage medium located in the memory 202. The processor 201 reads the executable instructions included in the software modules in the memory 202 and, in combination with necessary hardware (for example, including the processor 201 and other components connected to the bus 205), completes the text summary analysis method provided by the embodiment of the present invention.

[0112] As an example, the processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0113] As an example of a hardware implementation of the text summary analysis apparatus provided in an embodiment of the present invention, the apparatus provided in an embodiment of the present invention can be directly executed by a processor 201 in the form of a hardware decoding processor. For example, the text summary analysis method provided in an embodiment of the present invention can be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0114] The memory 202 in the embodiment of the present invention is used to store various types of data to support the operation of the electronic device 20. Examples of such data include any executable instructions for operating on the electronic device 20, such as executable instructions. The program implementing the text summary analysis method of the embodiment of the present invention may be included in the executable instructions.

[0115] In other embodiments, the text summary analysis device provided by the embodiment of the present invention can be implemented in software. Figure 2 The text summary analysis device 2020 stored in the memory 202 is shown. This device can be software in the form of a program or plug-in, and includes a series of modules. As an example of a program stored in the memory 202, the text summary analysis device 2020 can be included. The text summary analysis device 2020 includes the following software modules: an information transmission module 2081 and an information processing module 2082. When the software modules in the text summary analysis device 2020 are read into the RAM and executed by the processor 201, the text summary analysis method provided by the embodiment of the present invention is implemented. The functions of each software module in the text summary analysis device 2020 are described below:

[0116] The information transmission module 2081 is used to obtain the target text and summary text in the text summary analysis environment.

[0117] The information processing module 2082 is configured to process the summary text using a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text.

[0118] The information processing module 2082 is configured to process the target text using a text information processing model to determine pseudo reference information corresponding to the target text.

[0119] The information processing module 2082 is configured to determine a center weighting parameter that matches the target text based on the pseudo reference information corresponding to the target text.

[0120] The information processing module 2082 is configured to determine a relevance parameter between the target text and the summary text based on a self-reference redundancy parameter corresponding to the summary text and a center weighting parameter matching the target text.

[0121] The information processing module 2082 is configured to determine an analysis result of the summary text based on a relevance parameter between the target text and the summary text.

[0122] according to Figure 2 In one aspect of the electronic device shown, the present application further provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of the computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the various embodiments and combinations of embodiments provided in the various optional implementations of the above-mentioned text summary analysis method.

[0123] Combine Figure 2 The electronic device 20 shown illustrates the text summary analysis method provided by the embodiment of the present invention, see Figure 3 , Figure 3 This is an optional flow chart of the text summary analysis method provided by the embodiment of the present invention. It can be understood that: Figure 3 The steps shown can be performed by various electronic devices running the text summary analysis device, such as a dedicated terminal, electronic device or electronic device cluster with a retrieval instruction processing function. Figure 3 The steps shown are explained.

[0124] Step 301: The text summary analysis apparatus obtains a target text and a summary text in a text summary analysis environment.

[0125] In practical applications, in order to help users quickly understand the main idea of ​​a text message and decide whether to read the entire text message, different machine learning neural networks can usually generate corresponding text summaries, that is, a summary paragraph consisting of several sentences extracted from the text message, which is displayed at the beginning of the text message, or displayed in the text recommendation summary text of the text message. The quality of the summary texts generated by different neural networks can be evaluated, that is, whether they can accurately represent the entire text message, and then adjust the text summaries of poor quality or the corresponding neural networks accordingly to improve the reliability and efficiency of users in screening text information.

[0126] Since the types of target texts are different, the fields of text processing are also different. The content of text information in different fields varies greatly. Therefore, in order to improve the evaluation speed of summary text, the corresponding word segmentation library can be triggered according to the text parameter information carried by the target text; the target text is segmented by the word dictionary of the triggered word segmentation library to form different word-level feature vectors; the different word-level feature vectors are denoised to form a word-level feature vector set corresponding to the target text. In some embodiments of the present invention, a dynamic noise threshold that matches the use environment of the text information processing model can be determined; the different word-level feature vectors are denoised according to the dynamic noise threshold, and a dynamic word segmentation strategy that matches the dynamic noise threshold is triggered; the target text is segmented according to the dynamic word segmentation strategy that matches the dynamic noise threshold to form a dynamic word-level feature vector set corresponding to the target text. For example, in the use environment of academic translation, the text information displayed by the terminal only includes the text information of the academic paper. The dynamic noise threshold that matches the use environment of the text information processing model needs to be smaller than the dynamic noise threshold in the reading environment of entertainment information text.

[0127] Similarly, a fixed noise threshold corresponding to the use environment of the text information processing model can also be determined; the different word-level feature vectors are denoised according to the fixed noise threshold, and a fixed word segmentation strategy matching the fixed noise threshold is triggered; the target text is segmented according to the fixed word segmentation strategy matching the fixed noise threshold, and a fixed word-level feature vector set corresponding to the target text is generated. When the text information processing model is fixed in a corresponding hardware mechanism, such as an on-board terminal or a smart medical system, when the use environment is professional terminology text information (or text information in a certain field), since the noise is relatively simple, by fixing the fixed noise threshold corresponding to the text information processing model, the processing speed of the text information processing model can be effectively improved, the user's waiting time can be reduced, and the user's experience can be improved. Furthermore, since the text processed by the text information processing model includes not only text information in a single language but also complex text information in multiple languages ​​(for example, a mixed Chinese and English academic paper as text information), unlike English, which directly uses spaces as the interval between words, Chinese text needs to be segmented accordingly because words in Chinese can contain complete information. The Chinese word segmentation tool Jieba can be used to segment Chinese text. Furthermore, the keyword set after segmentation needs to be processed to remove stop words, as words like "yes" and "can" are not helpful for evaluating the corresponding summary text. For example, for the text "Yes, I like to do experiments", segmentation and removal of stop words will yield a set of two keywords, "like / do experiments" (using / as a delimiter, the same applies below). This can effectively improve the processing speed of the text information processing model.

[0128] Step 302: The text summary analysis device processes the summary text through a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text.

[0129] In some embodiments of the present invention, reference Figure 4 , Figure 4 The following is a schematic diagram of an optional data processing flow for the summary text evaluation process in an embodiment of the present invention, wherein the summary text is processed by a text information processing model to determine the self-reference redundancy parameter corresponding to the summary text, which can be achieved by the following method:

[0130] The summary text is processed by the encoding network of the text information processing model to extract the summary feature vector that matches the summary text; the first word-level latent vector corresponding to the summary text is determined based on the summary feature vector; the first word-level latent vector is subjected to maximum pooling processing to form a first sentence-level latent vector; the first word-level latent vector and the first sentence-level latent vector are spliced ​​to form a first spliced ​​vector; the similarity between different words in the first word-level latent vector and the similarity between different sentences in the first sentence-level latent vector are determined through the first spliced ​​vector; the self-reference redundancy parameter corresponding to the summary text is determined through the similarity between different words and the similarity between different sentences. In particular, since the summary texts generated by different machine learning neural network models are usually unlabeled information when evaluating them (for example, the real-time characteristics of the summary of medical records make it impossible to label the medical record information of each case), the self-reference redundancy parameter corresponding to the summary text can effectively determine the proportion of redundant information in the summary text. The higher the quality of the summary, the less redundant information it should contain. Hybrid representation based on summary Refer to Formula 1 to obtain the redundancy of the evaluation summary:

[0131]

[0132] where x i represents the i-th element in x, and “j:i≠j” means that the semantic similarity between the word or sentence and itself is not considered. red ∈[-1,1] represents the redundancy score. As can be seen from the formula, the redundancy score uses the abstract itself as a reference to calculate the semantic similarity of each word and sentence in the abstract with other words and sentences, and then averages the results to get the redundancy score of the entire abstract.

[0133] Step 303: The text summary analysis device processes the target text through a text information processing model to determine pseudo reference information corresponding to the target text.

[0134] In some embodiments of the present invention, processing the target text by a text information processing model to determine the pseudo reference information corresponding to the target text can be achieved by:

[0135] Determine the second word-level latent vector corresponding to the target text based on the summary feature vector; perform pooling on the second word-level latent vector to form a second sentence-level latent vector; determine the centrality score corresponding to each sentence in the target text based on the second sentence-level latent vector; sort the centrality score corresponding to each sentence in the target text, and determine the pseudo-reference information corresponding to the target text based on the sorting result of the centrality score. When screening sentences in different numbers of sorting results, the number of words in the pseudo-reference information can be determined based on the text summary analysis environment; based on the number of words in the pseudo-reference information, obtain a corresponding number of sentences in the sorting result of the centrality score as the pseudo-reference information corresponding to the target text. k Take the example of summary x, where the text processing model may include an encoder network and a decoder network. In some embodiments of the present invention, the text information processing model may be a bidirectional attention neural network model (BERTBidirectional Encoder Representations from Transformers). Figure 5 , Figure 5 This is an optional structural diagram of a text information processing model in an embodiment of the present invention, wherein the Encoder includes: N = 6 identical layers, each layer contains two fused sub-layer networks (sub-layers). The first sub-layer is a multi-head attention layer and then a simple fully connected layer. Each sub-layer has a residual connection and normalization. The decoder network (Decoder) includes: N = 6 identical layers, where the layer contains three fused sub-layer networks (sub-layers), one of which can be configured as a self-attention layer network (self-attention layer), and the last layer of the text information processing model can be configured as a fully connected layer network, through Figure 5 The text information processing model shown in the figure, the SUPERT text processing model converts the document d k The summary x is decomposed into sentences, and then the word segmenter and word embedding based on the BERT network are used to segment each sentence into a word sequence and encode each word into a hidden vector representation. The summary x can be represented by the hidden vector representation sequence corresponding to the word sequence, which is denoted as Where n represents the length of the summary, i.e. the number of words.k The hidden vector representation of each sentence in d is obtained by performing average pooling or maximum pooling on the hidden vector representation of all words in each sentence. Then the document d k The latent vector representation of each sentence is input into the PacSum algorithm to calculate the centrality score of each sentence. k Sort all the sentences in the document d, and the M sentences with the highest centrality scores form the document d k The pseudo-reference r. By linking the selected sentences in order, we can also get the latent vector representation sequence corresponding to the pseudo-reference word sequence: Where m represents the number of pseudo-reference words.

[0136] Step 304: The text summary analysis apparatus determines a center weighting parameter that matches the target text based on the pseudo reference information corresponding to the target text.

[0137] Combine Figure 2 The electronic device 20 shown illustrates the text summary analysis method provided by the embodiment of the present invention, see Figure 6 , Figure 6 This is an optional flowchart of text summary analysis provided by an embodiment of the present invention. It can be understood that: Figure 6 The steps shown can be performed by various electronic devices running a text summary analysis device, such as a dedicated terminal, server, or server cluster with a text summary analysis function, for training a text information processing model to determine model parameters that are compatible with the text information processing model. Specifically, the steps include:

[0138] Step 601: The server concatenates the second word-level latent vector and the second sentence-level latent vector to form a second concatenated vector.

[0139] Step 602: The server determines the centrality score corresponding to each sentence in the target text.

[0140] Step 603: The server normalizes the centrality score corresponding to each sentence in the target text to form a sentence-level centrality score.

[0141] Step 604: The server determines the sentence to which each word in the pseudo-reference information belongs, and determines a word-level centrality score based on the centrality score of the sentence to which each word in the pseudo-reference information belongs.

[0142] Step 605: The server determines the weight parameters corresponding to each sentence and word in the pseudo-reference information according to the sentence-level centrality score and the word-level centrality score.

[0143] The weight parameters corresponding to each sentence and word in the pseudo-reference information corresponding to the determined target text may be different, so as to achieve more accurate distinction of the importance of sentences and words in the pseudo-reference information.

[0144] Step 606: The server determines a center weighting parameter that matches the target text based on the second concatenation vector and the weight parameter corresponding to each sentence and word in the pseudo reference information.

[0145] In some embodiments of the present invention, determining the center weighting parameter for matching the target text based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information can be achieved by:

[0146] Based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information, the first splicing vector is used to determine the accuracy parameter corresponding to the target text; the first splicing vector is used to determine the recall parameter corresponding to the target text through the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information; and the center weight parameter matching the target text is determined based on the accuracy parameter and the recall parameter. Figure 7 , Figure 7 This is an optional data processing flow diagram of the summary text evaluation process in an embodiment of the present invention, wherein the latent vector corresponding to the word sequence of the summary text is represented by The latent vector representation corresponding to the pseudo-reference word sequence Then, the latent vector representation of all words in each sentence in the summary and pseudo-reference is maximized and pooled to obtain the sentence-level latent vector representation of the summary. Where N is the number of sentences in the summary and the sentence-level latent vector representation of the pseudo-reference Where M is the number of pseudo-reference sentences. Finally, the word-level latent vector representation and sentence-level latent vector representation are linked to obtain a hybrid representation to represent the summary and pseudo-reference. The hybrid representation of the summary is denoted as The mixed representation of pseudo-reference is denoted as

[0147] In some embodiments of the present invention, different weights can be assigned to the words and sentences of the pseudo-reference to show the difference in their importance and help the model calculate the relevance score of the summary more accurately. When establishing the pseudo-reference, the centrality score of each sentence of the pseudo-reference is retained, and then these scores are normalized and recorded as The superscript s represents the sentence level. Furthermore, we can refer to Formula 2 and Formula 3 to obtain the word-level centrality score through inheritance:

[0148]

[0149] In formula (5), “i:w j∈ s i " indicates that each pseudo-reference word w j From the statement to which it belongs i Inherit the same centrality score. Since the semantics of words constitute the semantics of sentences, this centrality score inheritance is reasonable. Finally, refer to Formula 4, Formula 5, and Formula 6 by linking the word-level centrality scores and sentence-level centrality scores And after normalization, we get the center score of the pseudo-reference mixed representation:

[0150]

[0151] Where "sum" represents the sum of all elements.

[0152] After obtaining the centrality scores of each word and sentence in the pseudo-reference, we can calculate the relevance score of the centrality-weighted summary. j represents the jth element in the summary mixed representation x, Represents pseudo-reference mixed representation R k The i-th element in a i represents the i-th element in the pseudo-reference centrality score a. Referring to Formula 7, Formula 8, and Formula 9, the centrality-weighted relevance score is calculated as follows:

[0153]

[0154] In some embodiments of the present invention, determining the relevance parameter between the target text and the summary text based on the self-referential redundancy parameter corresponding to the summary text and the center weighted parameter of the target text can be achieved by:

[0155] When the weight of the recall parameter corresponding to the target text in the text summary analysis environment is less than or equal to the weight of the accuracy parameter, the center weighted parameter matching the target text is used as the relevance parameter between the target text and the summary text; or, when the weight of the recall parameter corresponding to the target text in the text summary analysis environment is greater than the weight of the accuracy parameter, the ratio of the weight of the recall parameter to the weight of the accuracy parameter is determined; based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text, the relevance parameter between the target text and the summary text is determined by the ratio of the weight of the recall parameter to the weight of the accuracy parameter. Continue to refer to Figure 7 , the F1 score is used as the sum of the summary x and the document d k In addition to the score based on F1, in some embodiments of the present invention, the score based on F β The correlation score is calculated using the following method, where the physical meaning of β is that Recall is considered β times more important than Precision. Specifically, refer to Formula 10 and Formula 11 for the calculation method as follows:

[0156]

[0157] where |R k |=m+M represents R k The number of elements in the document, |X| = n + N, represents the number of elements in x. γ is an adjustable hyperparameter. β is adaptively determined based on the ratio of the number of elements in the mixed representation of the pseudo-reference and the summary. The longer the pseudo-reference and the shorter the summary, the larger β becomes. This setting is motivated by the fact that longer pseudo-references represent more important information in the document. However, shorter summaries are less likely to fully capture the important content of the pseudo-reference, resulting in lower recall. In this case, increasing the recall weight helps filter out such summaries.

[0158] Step 305: The text summary analysis apparatus determines a relevance parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text.

[0159] The final evaluation score of the summary represents the quality of the summary. Referring to Formula 12, the final evaluation score is a linear combination of the relevance score and the redundancy score:

[0160]

[0161] Where λ is an adjustable hyperparameter and score∈[-1,1] represents the final evaluation score.

[0162] Step 306: The text summary analysis device determines an analysis result of the summary text based on the relevance parameter between the target text and the summary text.

[0163] In some embodiments of the present invention, when the text information to be evaluated is composed of different texts (for example, the user's medical record information is a collection of text information of diagnosis and treatment results from multiple departments), for this summary text evaluation environment, it is necessary to calculate the summaries x and each document d. k The relevance scores are then averaged to obtain the final relevance score of the summary. Referring to Formula 13, the calculation method is as follows:

[0164]

[0165] “mean” means taking the average value. Represents the summary and document d k The correlation score, the subscript "*" or 1 or β indicates the corresponding F score calculation method, score rel ∈[-1, 1] represents the final relevance score.

[0166] The text summary analysis method provided by the present invention is described below by taking medical information in a medical diagnosis environment as an example. Figure 8 , Figure 8 The figure is a schematic diagram of processing different medical record text information. Among them, by extracting the corresponding summary text from different medical records of the same patient, doctors can obtain a more comprehensive understanding. In order to extract the summary text in the medical record information to assist doctors in diagnosis and treatment analysis, the text of the medical record information can be processed through the text information processing model. Due to the complex content of the medical record text information (many professional terms, many abbreviations of professional names, and long sentences), Figure 8Taking the medical record text information shown in the figure as an example, where T1MD represents type 1 diabetes and T2MD represents type 2 diabetes, although electronic medical records can collect some biometric information such as gender, age, blood test results, temperature, and blood pressure through patient vital signs and question-and-answer sessions, some machine learning methods only consider latent vector representations at the word level and ignore sentence-level latent vector representations during the text information processing model. This results in a single semantic mapping between pseudo-references and summaries, making it difficult to accurately display the relevance of the two. Furthermore, information in pseudo-references is equally weighted. In reality, the core information of a document is unevenly distributed across each sentence, resulting in relatively important and less important sentences within the pseudo-references. Simply assuming that all sentences are equally important can introduce additional noise into the relevance calculation. Consequently, the generated summary text may contain redundant information, hindering doctors' auxiliary diagnosis and treatment. For example, specific internal lesion information, such as hemorrhage or infarction lesions in the heart and brain, is often not available through the medical record summary text and can only be obtained by doctors reviewing the medical record information. Therefore, it is necessary to evaluate the summary text generated by different machine learning text information processing models.

[0167] To solve the above problems, refer to Figure 9 , Figure 9 An optional flowchart of the text summary analysis method provided in an embodiment of the present invention, wherein the user may be a doctor, and the target text information to be processed is different medical records of the same patient and summary texts of the medical records obtained by different machine learning neural networks. The method specifically includes the following steps:

[0168] Step 901: Obtain medical record information text and summary text generated by different neural network models.

[0169] Among them, such as Figure 8 As shown, medical record text information can include the patient's current medical history, including muscle weakness, sensory symptoms, aphasia, blurred vision, dizziness, headaches, nausea, vomiting, cognitive impairment, and impaired consciousness. Past medical history includes diabetes, hypertension, heart disease, hyperlipidemia, stroke, and aspirin use. Family history includes information about parents' history of cerebrovascular and cardiovascular disease, children's history of cerebrovascular and cardiovascular disease, and siblings' history of cerebrovascular and cardiovascular disease. Therefore, this information can be easily converted into numerical features, such as -1 for no, 1 for yes, and 0 for unclear. By evaluating summary text, a machine learning neural network tailored to the diagnostic and treatment environment can be selected to process large amounts of medical record information.

[0170] Step 902: Determine the self-reference redundancy parameters corresponding to different summary texts respectively.

[0171] Step 903: Determine pseudo reference information corresponding to the medical record information serving as the target text.

[0172] Step 904: Determine a centrality weighting parameter based on the pseudo reference information.

[0173] Step 905: Based on the self-reference redundancy parameters and centrality weighted parameters corresponding to different summary texts, the correlation parameters between the medical record information and different summary texts are determined by the ratio of the weight of the recall parameter to the weight of the precision parameter.

[0174] Step 906: Based on the correlation parameters between the medical record information and different summary texts, a text information processing model that matches the auxiliary diagnosis and treatment environment is selected.

[0175] Therefore, in different medical terminals, the summary text evaluation method provided in this application can automatically evaluate the summary texts generated by different neural network models, obtain a neural network model that is compatible with the diagnosis and treatment environment, and obtain the most suitable summary information to assist in medical diagnosis.

[0176] Beneficial technical effects:

[0177] The method obtains a target text and a summary text in a text summary analysis environment; processes the summary text through a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text; processes the target text through a text information processing model to determine pseudo-reference information corresponding to the target text; determines a center weighting parameter matching the target text based on the pseudo-reference information corresponding to the target text; determines a relevance parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighting parameter matching the target text; and determines an analysis result of the summary text based on the relevance parameter between the target text and the summary text. Thus, not only is automated evaluation of the summary text achieved, but also, through the center weighting parameter matching the target text, the influence of noise information in the pseudo-reference information on the analysis result of the summary text is reduced; and through the self-reference redundancy parameter corresponding to the summary text, summary texts without manually annotated references are more accurately automatically processed, thereby improving the efficiency of summary text evaluation and enhancing the user experience.

[0178] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A text summary analysis method, characterized in that: The method comprises: Get the target text and summary text; Processing the summary text using a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text, wherein the self-reference redundancy parameter is used to characterize a proportion of redundant information in the summary text; Processing the target text using the text information processing model to determine pseudo-reference information corresponding to the target text, wherein the pseudo-reference information is composed of important sentences in the target text, and the important sentences represent a preset number of sentences with the highest centrality scores in a ranking result of centrality scores of multiple sentences in the target text; Concatenate the second word-level latent vector and the second sentence-level latent vector to form a second concatenated vector; Determining a weight parameter corresponding to each sentence and word in the pseudo reference information corresponding to the target text; Determining a center weighting parameter that matches the target text based on the second splicing vector and the weight parameters corresponding to each sentence and word in the pseudo reference information; Determining a relevance parameter between the target text and the summary text based on a self-reference redundancy parameter corresponding to the summary text and a center weighted parameter matching the target text; An analysis result of the summary text is determined based on a relevance parameter between the target text and the summary text.

2. The method according to claim 1, characterized in that The step of processing the summary text by using a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text includes: Processing the summary text through an encoding network of a text information processing model to extract a summary feature vector that matches the summary text; Determining a first word-level latent vector corresponding to the summary text according to the summary feature vector; Performing maximum pooling on the first word-level latent vector to form a first sentence-level latent vector; concatenating the first word-level latent vector and the first sentence-level latent vector to form a first concatenated vector; Determining similarities between different words in the first word-level latent vector using the first concatenated vector; Determining similarities between different sentences in the first sentence-level latent vector using the first concatenated vector; The self-reference redundancy parameter corresponding to the summary text is determined according to the similarity between different words and the similarity between different sentences.

3. The method according to claim 2, characterized in that The processing of the target text by the text information processing model to determine pseudo reference information corresponding to the target text includes: Determining a second word-level latent vector corresponding to the target text according to the summary feature vector; Performing pooling processing on the second word-level latent vector to form a second sentence-level latent vector; Determining a centrality score corresponding to each sentence in the target text based on the second sentence-level latent vector; The centrality scores corresponding to each sentence in the target text are sorted, and the pseudo-reference information corresponding to the target text is determined according to the sorting results of the centrality scores.

4. The method according to claim 3, characterized in that The step of sorting the centrality scores corresponding to each sentence in the target text and determining the pseudo-reference information corresponding to the target text according to the sorting results of the centrality scores includes: determining the number of words in the pseudo-reference information according to a text summary analysis environment; Based on the number of words in the pseudo-reference information, a corresponding number of sentences are obtained from the ranking result of the centrality scores as pseudo-reference information corresponding to the target text.

5. The method according to claim 1, wherein The determining of the weight parameters corresponding to each sentence and word in the pseudo reference information corresponding to the target text includes: Determining a centrality score corresponding to each sentence in the target text; Normalizing the centrality score corresponding to each sentence in the target text to form a sentence-level centrality score; Determining the sentence to which each word in the pseudo-reference information belongs; Determining a word-level centrality score based on the centrality score of the sentence to which each word in the pseudo-reference information belongs; The weight parameters corresponding to each sentence and word in the pseudo-reference information are determined according to the sentence-level centrality score and the word-level centrality score.

6. The method according to claim 2, characterized in that The determining of a center weighting parameter matching the target text based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information includes: Determining an accuracy parameter corresponding to the target text using the first splicing vector based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information; Determine a recall parameter corresponding to the target text using the first splicing vector by using the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo reference information; A center weighting parameter that matches the target text is determined according to the accuracy parameter and the recall parameter.

7. The method according to claim 6, characterized in that The determining of the relevance parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text includes: When the weight of the recall parameter corresponding to the target text in the text summary analysis environment is less than or equal to the weight of the accuracy parameter, the center weighted parameter matching the target text is used as the relevance parameter between the target text and the summary text.

8. The method according to claim 6, characterized in that The determining of the relevance parameter between the target text and the summary text based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text includes: When the weight of the recall parameter corresponding to the target text in the text summary analysis environment is greater than the weight of the accuracy parameter, Determining a ratio of the weight of the recall parameter to the weight of the precision parameter; Based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text, the relevance parameter between the target text and the summary text is determined by the ratio of the weight of the recall parameter to the weight of the accuracy parameter.

9. The method according to claim 1, characterized in that The method further comprises: Triggering the corresponding word segmentation library according to the text parameter information carried by the target text; Perform word segmentation processing on the target text through the triggered word segmentation library dictionary to form different word-level feature vectors; The different word-level feature vectors are subjected to noise removal processing to form a word-level feature vector set corresponding to the target text.

10. The method according to claim 9, characterized in that The performing denoising on the different word-level feature vectors to form a word-level feature vector set corresponding to the target text includes: Determining a dynamic noise threshold that matches the usage environment of the text information processing model; Performing noise removal on the different word-level feature vectors according to the dynamic noise threshold, and triggering a dynamic word segmentation strategy that matches the dynamic noise threshold; The target text is segmented according to a dynamic word segmentation strategy that matches the dynamic noise threshold, so as to form a set of dynamic word-level feature vectors corresponding to the target text.

11. The method according to claim 10, characterized in that The performing denoising on the different word-level feature vectors to form a word-level feature vector set corresponding to the target text includes: Determining a fixed noise threshold corresponding to the usage environment of the text information processing model; Performing denoising on the different word-level feature vectors according to the fixed noise threshold, and triggering a fixed word segmentation strategy that matches the fixed noise threshold; According to a fixed word segmentation strategy that matches the fixed noise threshold, the target text is segmented, and a fixed word-level feature vector set corresponding to the target text is obtained.

12. A text summary analysis device, characterized in that: The device comprises: An information transmission module is used to obtain the target text and summary text in the text summary analysis environment; an information processing module, configured to process the summary text using a text information processing model to determine a self-reference redundancy parameter corresponding to the summary text, wherein the self-reference redundancy parameter is used to characterize a proportion of redundant information in the summary text; The information processing module is configured to process the target text using a text information processing model to determine pseudo-reference information corresponding to the target text, wherein the pseudo-reference information is composed of important sentences in the target text, and the important sentences represent a preset number of sentences with the highest centrality scores in a ranking result of centrality scores of multiple sentences in the target text; The information processing module is configured to concatenate the second word-level latent vector and the second sentence-level latent vector to form a second concatenated vector; determine a weight parameter corresponding to each sentence and word in the pseudo-reference information corresponding to the target text; and determine a center weighting parameter that matches the target text based on the second concatenated vector and the weight parameter corresponding to each sentence and word in the pseudo-reference information; The information processing module is configured to determine a relevance parameter between the target text and the summary text based on a self-reference redundancy parameter corresponding to the summary text and a center weighting parameter matching the target text; The information processing module is configured to determine an analysis result of the summary text based on a relevance parameter between the target text and the summary text.

13. The device according to claim 12, characterized in that The information processing module is further configured to process the summary text through an encoding network of a text information processing model to extract a summary feature vector matching the summary text; determine a first word-level latent vector corresponding to the summary text based on the summary feature vector; perform maximum pooling processing on the first word-level latent vector to form a first sentence-level latent vector; concatenate the first word-level latent vector and the first sentence-level latent vector to form a first concatenated vector; determine, through the first concatenated vector, the similarity between different words in the first word-level latent vector and the similarity between different sentences in the first sentence-level latent vector; and determine a self-reference redundancy parameter corresponding to the summary text based on the similarity between different words and the similarity between different sentences.

14. The device according to claim 13, characterized in that The information processing module is further configured to determine a second word-level latent vector corresponding to the target text based on the summary feature vector; perform pooling processing on the second word-level latent vector to form a second sentence-level latent vector; and determine a centrality score corresponding to each sentence in the target text based on the second sentence-level latent vector; The centrality scores corresponding to each sentence in the target text are sorted, and the pseudo-reference information corresponding to the target text is determined according to the sorting results of the centrality scores.

15. The device according to claim 14, characterized in that The information processing module is further configured to determine the number of words in the pseudo-reference information according to the text summary analysis environment; and based on the number of words in the pseudo-reference information, obtain a corresponding number of sentences in the ranking result of the centrality score as the pseudo-reference information corresponding to the target text.

16. The device according to claim 12, characterized in that The information processing module is further configured to determine a centrality score corresponding to each sentence in the target text; normalize the centrality score corresponding to each sentence in the target text to form a sentence-level centrality score; determine the sentence to which each word in the pseudo-reference information belongs; and determine a word-level centrality score based on the centrality score of the sentence to which each word in the pseudo-reference information belongs. The weight parameters corresponding to each sentence and word in the pseudo-reference information are determined according to the sentence-level centrality score and the word-level centrality score.

17. The device according to claim 13, characterized in that The information processing module is further used to determine the accuracy parameter corresponding to the target text using the first splicing vector based on the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo-reference information; determine the recall parameter corresponding to the target text using the first splicing vector through the second splicing vector and the weight parameter corresponding to each sentence and word in the pseudo-reference information; and determine the center weighting parameter matching the target text based on the accuracy parameter and the recall parameter.

18. The device according to claim 17, characterized in that The information processing module is further configured to, when the weight of the recall parameter corresponding to the target text in the text summary analysis environment is less than or equal to the weight of the accuracy parameter, use the center weighted parameter that matches the target text as the relevance parameter between the target text and the summary text.

19. The device according to claim 17, characterized in that The information processing module is further configured to determine a ratio of the weight of the recall parameter to the weight of the accuracy parameter when the weight of the recall parameter corresponding to the target text in the text summary analysis environment is greater than the weight of the accuracy parameter; Based on the self-reference redundancy parameter corresponding to the summary text and the center weighted parameter matching the target text, the relevance parameter between the target text and the summary text is determined by the ratio of the weight of the recall parameter to the weight of the accuracy parameter.

20. The device according to claim 12, characterized in that The information processing module is further used to trigger the corresponding word segmentation library according to the text parameter information carried by the target text; perform word segmentation processing on the target text through the word dictionary of the triggered word segmentation library to form different word-level feature vectors; and perform noise removal processing on the different word-level feature vectors to form a set of word-level feature vectors corresponding to the target text.

21. The device according to claim 20, characterized in that The information processing module is further used to determine a dynamic noise threshold that matches the usage environment of the text information processing model; perform noise removal processing on the different word-level feature vectors according to the dynamic noise threshold, and trigger a dynamic word segmentation strategy that matches the dynamic noise threshold; perform word segmentation processing on the target text according to the dynamic word segmentation strategy that matches the dynamic noise threshold, and form a set of dynamic word-level feature vectors corresponding to the target text.

22. The device according to claim 21, characterized in that The information processing module is further used to determine a fixed noise threshold corresponding to the usage environment of the text information processing model; perform denoising on the different word-level feature vectors according to the fixed noise threshold, and trigger a fixed word segmentation strategy that matches the fixed noise threshold; perform word segmentation on the target text according to the fixed word segmentation strategy that matches the fixed noise threshold, and obtain a set of fixed word-level feature vectors corresponding to the target text.

23. An electronic device, characterized in that: The electronic device comprises: a memory for storing executable instructions; The processor is configured to implement the text summary analysis method according to any one of claims 1 to 11 when running the executable instructions stored in the memory.

24. A computer-readable storage medium storing executable instructions, characterized in that: When the executable instructions are executed by a processor, the text summary analysis method according to any one of claims 1 to 11 is implemented.

25. A computer program product storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the text summary analysis method according to any one of claims 1 to 11 is implemented.

Images