A summary determination method and related device

By calculating the summary weight based on the importance and redundancy parameters in the text set and filtering the target text fragments, the problem of low generation efficiency in the prior art is solved, and the effect of rapid generation of summary is achieved.

CN112328783BActive Publication Date: 2024-11-08TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011333789.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-11-24
Publication Date
2024-11-08
Estimated Expiration
2040-11-24

AI Technical Summary

Technical Problem

The existing automatic digest technology has insufficient generation efficiency, especially when processing long original documents or multiple documents, the calculation is high, resulting in slower generation of summary content.

Method used

By calculating the summary weight based on the importance parameters and redundancy parameters of the text fragments in the text collection, filter out more important and low redundancy target text fragments, reduce the number of text fragments required to generate the summary content, and improve the generation efficiency.

Benefits of technology

It effectively reduces the number of text fragments required to generate the summary content, improves the efficiency of the summary content generation, and enables the rapid generation of the summary content to be generated in a short period of time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112328783B_ABST
    Figure CN112328783B_ABST
Patent Text Reader

Abstract

The embodiment of the present application discloses a summary determination method and related devices, which at least involve natural language processing and machine learning in artificial intelligence. For a text set including at least one text to be processed, in the process of determining the summary content corresponding to the text set, the summary weight of the text segment is obtained based on the importance parameter of the text segment included in the text to be processed relative to the text to be processed, and the redundancy parameter relative to the determined part in the summary content. According to the summary weight, a relatively important text segment with low redundancy with the determined part of the current summary content is screened out from the text segments included in the text set, and the text segment is used as the target text segment, and the summary content is continuously generated through a small number of target text segments instead of all text segments, thereby reducing the number of text segments required to be processed to generate the summary content, reducing the amount of calculation, and improving the efficiency of generating the summary content.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a summary determination method and related devices. Background Art

[0002] The abstract can reflect the important information in the original document, and its length is generally much smaller than the original document. Users can understand the purpose of the main idea of ​​the original document by reading the abstract, so that they can quickly find documents that meet their needs from a large number of original documents, avoiding the screening time of reading long content.

[0003] In the related art, automatic summarization technology can be used to understand the content of the original document based on semantics according to a neural network model, and to automatically extract important information from the original document to generate a corresponding summary.

[0004] However, the above summary extraction method has limited application scope and has the problem of low generation efficiency in many scenarios. Summary of the invention

[0005] In order to solve the above technical problems, the present application provides a summary determination method and related devices, which can improve the efficiency of generating summaries.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] In one aspect, an embodiment of the present application provides a method for determining an abstract, the method comprising:

[0008] Acquire a text set including at least one text to be processed, wherein the text to be processed includes a plurality of text segments;

[0009] In the process of determining the summary content corresponding to the text set, the summary weight of the text segment is obtained by using the importance parameter of the text segment relative to the text to be processed and the redundancy parameter of the text segment relative to the determined part of the summary content;

[0010] Determining target text segments from the text segments included in the text collection based on the summary weights, wherein the number of the target text segments is less than the number of the text segments in the text collection;

[0011] The summary content corresponding to the text collection is generated according to the target text segment.

[0012] On the other hand, an embodiment of the present application provides a summary determination device, the device comprising: an acquisition unit, a first determination unit, a second determination unit, and a generation unit;

[0013] The acquisition unit is used to acquire a text set including at least one text to be processed, wherein the text to be processed includes a plurality of text segments;

[0014] The first determining unit is used to obtain the summary weight of the text segment by using the importance parameter of the text segment relative to the text to be processed and the redundancy parameter of the text segment relative to the determined part of the summary content in the process of determining the summary content corresponding to the text set;

[0015] The second determining unit is used to determine target text segments from the text segments included in the text collection based on the summary weight, the number of the target text segments being less than the number of the text segments in the text collection;

[0016] The generating unit is used to generate the summary content corresponding to the text set according to the target text segment.

[0017] In another aspect, the present application provides a computer device, the device comprising a processor and a memory:

[0018] The memory is used to store program code and transmit the program code to the processor;

[0019] The processor is used to execute the method described in the above aspects according to the instructions in the program code.

[0020] On the other hand, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method described in the above aspects.

[0021] On the other hand, an embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method described in the above aspects.

[0022] It can be seen from the above technical solution that for a text set including at least one text to be processed, in the process of determining the summary content corresponding to the text set, the summary weight of the text segment is obtained based on the importance parameter of the text segment included in the text to be processed relative to the text to be processed, and the redundancy parameter relative to the determined part in the summary content. Among them, the redundancy parameter is the similarity of the text segment to the determined part of the current summary content when the summary content is determined this time. Through the redundancy parameter, it can be determined which text segments in the text set have a high similarity with the determined part of the current summary content, and which text segments have a low similarity. Since the text segments with a high similarity based on the determined part of the current summary content are not very helpful for continuing to generate the summary content, therefore, according to the summary weight determined based on the importance parameter and the redundancy parameter, the more important text segments included in the text set and the low redundancy with the determined part of the current summary content can be screened out as the target text segments, and the summary content can be continued to be generated through a small number of target text segments. In this way, the number of text segments required to be processed to generate the summary content can be reduced, and the efficiency of generating the summary content can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 A schematic diagram of an application scenario of a summary determination method provided in an embodiment of the present application;

[0025] Figure 2 A schematic diagram of a process for determining a summary provided in an embodiment of the present application;

[0026] Figure 3 A schematic diagram of an application scenario of a summary determination method provided in an embodiment of the present application;

[0027] Figure 4 A schematic diagram of the structure of a summary determination device provided in an embodiment of the present application;

[0028] Figure 5 A schematic diagram of the structure of a server provided in an embodiment of the present application;

[0029] Figure 6 A schematic diagram of the structure of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The embodiments of the present application are described below in conjunction with the accompanying drawings.

[0031] When the original document is long or the user needs to read multiple original documents in a short time, the neural network model will generate corresponding summary content based on the entire content of the original document based on semantic understanding of the original document. Since long original documents or multiple original documents include more content, the amount of calculation is large and the summary content is generated slowly.

[0032] Based on this, the present application provides a summary determination method and related devices, which screen text fragments in a text collection through importance parameters and redundancy parameters, thereby generating corresponding summary content based on a smaller number of target text fragments, reducing the number of text fragments required to be processed to generate summary content, and improving the efficiency of summarizing content generation.

[0033] The abstract determination method provided in the embodiment of the present application is based on artificial intelligence. Artificial Intelligence (AI) is a 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. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that the machines have the functions of perception, reasoning and decision-making.

[0034] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0035] In the embodiments of the present application, the artificial intelligence software technologies mainly involved include the above-mentioned natural language processing, machine learning / deep learning and other directions. For example, it may involve semantic understanding (Semantic understanding) in natural language processing (NLP), robot question answering, including question understanding (query understanding), answer extraction (Answer extraction), answer argumentation (Answer argument), and it may also involve deep learning (Deep Learning) in machine learning (Machine learning, ML), including various types of artificial neural networks (Artificial Neural Network, ANN).

[0036] The summary determination method provided in this application can be applied to summary determination devices with data processing capabilities, such as terminal devices and servers. Specifically, the terminal device can be a smart phone, a desktop computer, a laptop computer, a tablet computer, a smart speaker, a smart watch, etc., but is not limited thereto; the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application does not limit this.

[0037] The abstract determines that the device can have the ability to implement natural language processing. Natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language used by people in daily life, so it is closely related to the study of linguistics. Natural language processing technology generally includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies. In an embodiment of the present application, the text processing device can process the text through text preprocessing, semantic understanding and other technologies in natural language processing.

[0038] This summary confirms that the device can have machine learning capabilities. Machine learning is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks.

[0039] In the summary determination method provided in the embodiment of the present application, the artificial intelligence model adopted mainly involves the application of machine learning, and redundancy parameters and importance parameters are determined through machine learning so as to screen out a small number of target text fragments from a text collection to generate corresponding summary content.

[0040] In addition, the text processing device provided in the embodiment of the present application also has cloud computing capabilities. Cloud computing refers to the delivery and use model of IT infrastructure, which means obtaining the required resources through the network in an on-demand, easily scalable manner; in a broad sense, cloud computing refers to the delivery and use model of services, which means obtaining the required services through the network in an on-demand, easily scalable manner. This service can be IT and software, Internet-related, or other services. Cloud computing is the product of the development and integration of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing.

[0041] With the development of the Internet, real-time data streams, and the diversification of connected devices, as well as the demand for search services, social networks, mobile commerce, and open collaboration, cloud computing has developed rapidly. Different from the previous parallel distributed computing, the emergence of cloud computing will promote revolutionary changes in the entire Internet model and enterprise management model from a conceptual perspective.

[0042] In an embodiment of the present application, the summary determination device may utilize cloud computing technology to obtain a text set including at least one text to be processed, so as to generate summary content corresponding to the text set.

[0043] To facilitate understanding of the technical solution of the present application, the summary determination method provided in the embodiment of the present application is introduced below with a server as a summary determination device in combination with actual application scenarios.

[0044] See also Figure 1 , Figure 1 A schematic diagram of an application scenario of a summary determination method provided in an embodiment of the present application. Figure 1 The application scenario shown includes a server 100 for generating summary content corresponding to a text collection.

[0045] The server 100 obtains a text collection, which includes at least one text to be processed, and each text to be processed includes a plurality of text segments. Figure 1 In the scenario shown, the text set obtained by the server 100 includes two texts to be processed, namely, text to be processed I and text to be processed II, wherein text to be processed I includes text segment A and text segment B, and text to be processed II includes text segment C and text segment D.

[0046] In the process of determining the summary content corresponding to the text set, the summary weight of the text segment is obtained based on the importance parameter of the text segment included in the to-be-processed text relative to the to-be-processed text, and the redundancy parameter relative to the determined part in the summary content. Among them, the importance parameter reflects the degree of relevance of the text segment to the to-be-processed text in which it is located, and the more important information in the to-be-processed text is determined by the importance parameter. The redundancy parameter is the degree of similarity of the text segment to the determined part of the current summary content when determining the summary content this time, and the more similar information in the text set can be determined by the redundancy parameter.

[0047] exist Figure 1In the scenario shown, multiple iterations are performed to generate summary content corresponding to the text set, and each iteration can generate a part of the summary content. In this embodiment, the last two iterations (t-th iteration and t+1-th iteration) of determining the summary content corresponding to the text set are used as an example for explanation. In the t-th iteration, the summary weight is determined by the importance parameter and the redundancy parameter, and the target text segment is determined from the text segments included in the text set according to the summary weight. For example, the summary weights of text segments A and text segments C in the text set are relatively high, so the target text segments include text segments A and text segments C. A part of the summary content can be generated through the target text segments, thereby obtaining the corresponding summary content. In the t+1th iteration, the redundancy parameter is re-determined based on the determined part of the current summary content, i.e., the summary content determined in the tth iteration, and the summary weight is further re-determined. According to the re-determined summary weight, it is determined that the redundancy of text fragments A and text fragment B is higher than the determined part of the current summary content. Since text fragments with higher similarity are not very helpful for continuing to generate summary content, the target text fragments are determined this time to include text fragments C and text fragment D. A part of the summary content can be generated through this target text fragment, and the summary content corresponding to the text set is generated in combination with the summary content determined in the tth iteration.

[0048] Based on the above, it can be known that, by determining the summary weight based on the importance parameter and the redundancy parameter, the text fragments included in the text set that are relatively important and have low redundancy with the determined part of the current summary content are screened out as target text fragments. Therefore, it is not necessary to generate the summary content corresponding to the text set with all the text fragments in the text set, and the summary content corresponding to the text set can be generated based on only a small number of target text fragments. Therefore, the number of text fragments that need to be processed to generate the summary content can be reduced, and the generation efficiency of the summary content can be improved.

[0049] In conjunction with the accompanying drawings, a method for determining an abstract provided in an embodiment of the present application is introduced below by taking a terminal device as a text processing device.

[0050] See also Figure 2 , which is a flow chart of a method for determining a summary provided in an embodiment of the present application. Figure 2 As shown, the summary determination method includes the following steps:

[0051] S201: Acquire a text set including at least one text to be processed.

[0052] By using a short summary that covers important information of the text, users can quickly determine whether it is necessary to read the entire text, or help users quickly filter out the required text. However, not all texts have pre-summarized summary content, such as news articles, product manuals, etc., so it is necessary to quickly extract the summary content corresponding to the text through the automatic summary generation method provided in the embodiment of the present application.

[0053] In this step, the text collection obtained by the terminal device is an object for which summary content needs to be generated, and the text collection includes at least one text to be processed. For example, the text collection includes multiple articles, and a corresponding summary content needs to be generated for the multiple articles. It should be noted that the embodiment of the present application is also applicable to the summary determination of a text collection that only includes one text to be processed.

[0054] Among them, a text to be processed can be divided into multiple text segments. Based on different division granularities, the text segments can be data structures such as paragraphs, sentences, and words. Taking the text segments as sentences as an example, the multiple text segments included in the text to be processed can be determined based on the separators in the text to be processed, such as periods.

[0055] S202: In the process of determining the summary content corresponding to the text set, the summary weight of the text segment is obtained by using an importance parameter of the text segment relative to the text to be processed and a redundancy parameter of the text segment relative to a determined part of the summary content.

[0056] The generation of summary content is generally not based on a one-time text processing, but in a process of multiple iterations, in which part of the summary content is generated based on each iteration, and finally the completed summary content is formed. Therefore, the summary content is gradually completed based on a time-series process.

[0057] In the related art, corresponding summary content is generated based on all text fragments included in the text collection. However, in many cases, the number of text fragments is actually very large. If all text fragments are used to generate summary content in each iteration, it will bring a high computational burden, resulting in a long time-consuming and low efficiency in summarizing the content.

[0058] However, since the summary content is an outline of the main content embodied in the text collection, not all text fragments in the text collection are helpful for generating summary content. Therefore, the text fragments included in the text collection can be screened to select text fragments that are helpful for generating summary content. The factors for selecting text fragments that are helpful for generating summary content from the text collection are specifically described below.

[0059] Factor 1: Importance parameter.

[0060] Since the summary content mainly reflects the important information of the text collection, the higher the relevance of the text fragment to the to-be-processed text (hereinafter referred to as the to-be-processed text), the more important information the text fragment reflects, and the more helpful it is for generating the summary content. Therefore, the text fragments can be screened based on the importance parameters of the text fragments relative to the to-be-processed text, and the importance parameters can be used to determine which text fragments in the to-be-processed text are more important and which text fragments are less important, so that the text fragments that are helpful for generating the summary content can be screened out in the text collection based on the importance parameters.

[0061] The following is a method for obtaining importance parameters, see formula (1):

[0062] r1 i =f([sl i ;idx1 i ;idx2 i ; l i ; simd i ]) (1)

[0063] Among them, the text set includes I text fragments, r1 i represents the importance parameter of the i-th text segment in the text set relative to the text to be processed, i∈I, i and I are both integers, I>1; sl i Indicates the length of the i-th text segment; idx1 i Indicates the absolute position of the i-th text segment in the text to be processed; idx2 i Indicates the relative position of the i-th text segment in the text to be processed; l i Represents the segment features of the i-th text segment; simd i Represents the cosine distance between the segment feature of the i-th text segment and the segment feature of the text to be processed.

[0064] Specifically, the above parameters are described in detail below.

[0065] Among them, f() means passing the parameter sl i ,idx1 i ,idx2 i , l i and simd i Get r1 i The present embodiment does not specifically limit the method for obtaining the importance parameter r1. i For example, r1 i It can be obtained through regression analysis of the Support Vector Machine (SVR) model.

[0066] The absolute position refers to the position of the text segment in the text to be processed. i It is unique. The absolute position can be used to clearly identify the specific position of the text segment in the text to be processed. For example, the third text segment in an article can be represented as ind13=3.

[0067] Relative position refers to the different positions of text fragments due to different ways of dividing the text to be processed. The relative positions of text fragments with similar positions are the same or have a small difference. i Different values ​​may correspond to different division methods. For example, when the text to be processed is evenly divided into n parts in order, the relative positions of the text fragments belonging to a part are the same, and the relative positions of each part are 0, 1, 2, ..., n-1. When the third text fragment in an article belongs to the second part, it can be expressed as ind23 = 2, and when the third text fragment in an article belongs to the third part, it can be expressed as ind23 = 3. The relative position is used to emphasize the approximate position of the text fragment in the head, middle, and tail of the text to be processed. Text fragments with different relative positions have different degrees of help to the summary content. For example, the text fragment at the head of the news is generally more important.

[0068] The embodiment of the present application provides an optional method for determining the segment features of a text segment, by taking the feature obtained from the last hidden state in the encoding process as l i For example, when a text segment encodes the target text using a query-based selective encoder (QBS-Encoder) algorithm based on user intent, the feature obtained by the last hidden state of the bidirectional long short-term memory (BiLSTM) is the segment feature l i QBS-Encoder and BiLSTM will be described later and will not be described here in detail.

[0069] After the text set is determined, the importance parameters of each text fragment will not change with the process of determining the summary and are relatively fixed parameters. Therefore, in the embodiment of the present application, the determined importance parameters can be recorded and directly called during the summary determination process without the need for repeated calculations.

[0070] Factor 2: Redundancy parameter.

[0071] In addition, since the determination of the summary content is a gradual and sequential completion of the content, in this scenario, not all text fragments with higher importance parameters are helpful in generating the summary content in each iteration, and it is necessary to further consider whether the content between the text fragment and the currently determined part of the summary content is repeated. For example, when the content of a text fragment has been refined into the current summary content or has been reflected in the current summary content during the previous iteration, the text fragment may actually be of little help in continuing to complete the summary content. For another example, if the importance parameter of a text fragment is high and the redundancy parameter is low, then there is a high possibility that this text fragment will be used to generate part of the summary in the summary content.

[0072] Therefore, the embodiment of the present application also considers the redundancy parameter of the text fragment relative to the determined part of the summary content. Among them, the redundancy parameter is the similarity of the text fragment to the determined part of the current summary content when the summary content is determined this time. Through the redundancy parameter, it can be determined which text fragments in the text collection have a high similarity with the determined part of the current summary content, and which text fragments have a low similarity. Using text fragments with a high similarity will make the generated summary content redundant, that is, the text fragment is not very helpful in generating the summary content, so that the text fragments that are helpful in generating the summary content can be screened out in the text collection based on the redundancy parameter.

[0073] For example, when the summary content corresponding to a text set is iteratively determined through multiple time steps, each time step can represent an iterative process, and each iterative process can generate a part of the summary content based on the text fragments in the text set, thereby generating the summary content corresponding to the text set according to multiple iterative processes.

[0074] The following takes the t-th time step among multiple time steps as an example, where t is an integer greater than or equal to 1. When determining the summary content at the t-th time step, the determined part of the current summary content is the partial summary determined at the previous t-1 time step, and the redundancy parameter of the text segment at the t-th time step is determined. See formula (2):

[0075]

[0076] Among them, r3 i represents the redundancy parameter of the i-th text segment relative to the determined part of the summary content; S represents the determined part of the current summary content. The set includes J summary text segments, and its division granularity can be consistent with the division granularity of the text segments. For example, the summary text segment and the text segments of the text set are both sentences; s j represents the jth summary text segment in S; s irepresents the i-th text segment; ROUGE-L is a way to evaluate the content of the summary. It can be understood that when t=1, S is empty, and the corresponding r3 is determined i A value of 0 means there is no redundancy.

[0077] Specifically, the above parameters are described in detail below.

[0078] ROUGE-L is a set of indicators that use the longest common subsequence to evaluate summary content in the Rouge evaluation method. The full name of Rouge is Recall-Oriented Understudy for Gisting Evaluation. The automatically generated summary content is compared with a set of reference summaries (usually manually generated) to obtain the corresponding score to measure the similarity between the automatically generated summary content and the reference summaries. L represents the longest common subsequence (LCS).

[0079] Based on the above, it can be known that the summary weight can be determined based on the importance parameter and the redundancy parameter. According to the role of the importance parameter and the redundancy parameter in determining the summary content, it can be clearly seen that the importance parameter is positively correlated with the summary weight, and the redundancy parameter is negatively correlated with the summary weight. For example, the summary weight can be referred to in formula (3):

[0080] imp i =α×r1 i -β×r3 i (3)

[0081] Among them, imp i represents the summary weight of the i-th text segment; α and β are weight adjustment parameters, for example, α+β=1.

[0082] S203: Determine a target text segment from the text segments included in the text set based on the summary weight.

[0083] Since the summary weight is determined based on the importance parameter and the redundancy parameter, the text fragments with high importance and low redundancy with the determined part of the current summary content can be screened out from the text fragments included in the text set according to the summary weight as the target text fragments. Among them, the number of target text fragments is less than the number of text fragments included in the text set, so that the summary content is generated based on the smaller number of target text fragments, reducing the calculation amount of generating the summary and improving the generation efficiency.

[0084] When the summary content corresponding to the text set is determined through multiple time steps, since each time step can generate a part of the summary content, the redundancy parameter corresponding to the text segment will change at each time step due to the change of the determined part of the current summary content. Therefore, the redundancy parameter corresponding to the text segment will be re-determined at each time step to update the corresponding summary weight, and then the target text segment required for this iterative processing will be re-determined based on the updated summary weight.

[0085] Continuing with the t-th time step among multiple time steps as an example, based on the determined part of the summary content obtained in the previous t-1 time step, the redundancy parameter of the text fragment in the t-th time step is re-determined, and the summary weight of the text fragment in the t-th time step is determined in combination with the importance parameter, and the target text fragment for the t-th time step is determined from the text fragments included in the text set based on the summary weight.

[0086] In order to select a target text segment that is more helpful in generating summary content, the top K text segments with the largest summary weights may be selected from the text segments included in the text set as the target text segments based on the summary weights.

[0087] Optionally, in an embodiment of the present application, when a text set includes multiple texts to be processed, the target text segment is selected based on the text set rather than a single text to be processed, and the aforementioned topK text segments are obtained based on the summary weight sorting of the text segments in the text set. For example, when a text set includes multiple texts to be processed, the text segments respectively included in the multiple texts to be processed, that is, all the text segments included in the text set, are sorted according to the summary weights, and the top K text segments with the largest summary weights are determined as the target text segments.

[0088] S204: Generate the summary content corresponding to the text set according to the target text segment.

[0089] The target text fragment is a text fragment that is relatively important in the text collection and has low redundancy with the determined part of the current summary content. Therefore, there is no need to generate the summary content corresponding to the text collection based on all the text fragments included in the text collection. The summary content corresponding to the text collection can be generated by a small number of target text fragments, which can reduce the number of text fragments that need to be processed to generate the summary content and improve the efficiency of generating the summary content.

[0090] When generating summary content corresponding to a text set through a target text segment, a text set feature corresponding to the text set can be generated based on the target text segment, and then a corresponding summary can be generated based on the text set feature. Continuing to take the t-th time step among multiple time steps as an example, after determining the target text segment for the t-th time step, the text set feature corresponding to the text set is generated through the target text segment of the t-th time step.

[0091] From the above, it can be seen that the redundancy parameters corresponding to the text fragment will be re-determined at each time step, so as to determine the summary weight and then determine the target text fragment. That is to say, the target text fragment determined at each time step may be different, and the text set features generated based on the target text fragment at each time step may be different, so the determined summary content is also different, which is explained in detail below.

[0092] According to the summary weight of the target text segment at the t-th time step, when generating the text set features corresponding to the text set, the attention weight corresponding to the target text segment at the t-th time step is adjusted, see formula (4):

[0093]

[0094] in, represents the adjusted attention weight corresponding to the i-th text segment at the t-th time step; represents the attention weight corresponding to the i-th text segment at the t-th time step; imp i represents the summary weight of the i-th text segment; SK represents the target text segment set consisting of the target text segments, s i Represents the i-th text segment in the text collection.

[0095] At the tth time step, when the i-th text segment in the text set belongs to the text segment in the target text segment set SK, the attention weight corresponding to the i-th target text segment is adjusted to When the i-th text segment in the text set does not belong to the text segments in the target text segment set SK, the attention weight corresponding to the i-th target text segment is adjusted to 0.

[0096] It is understandable that the corresponding attention weight can be obtained when determining the target text segment at the t-th time step. For example, when the decoder obtains the summary content, the decoder will also generate the attention weight of the target text segment.

[0097] The text set features corresponding to the text set are generated by adjusting the attention weight and the segment features of the target text segment at the t-th time step. See formula (5):

[0098]

[0099] in, represents the text set features corresponding to the text set; the target text segment set SK includes K target text segments, represents the adjusted attention weight corresponding to the k-th target text segment at the t-th time step, k∈K, k and K are both integers, K>0; l k represents the kth target text segment s in the target text segment set SK k The segment features of the target text segment s k With text snippets i are all text fragments in the text set, and the target text fragment s k The number of text fragments is less than s i The number of

[0100] The text set features thus obtained not only consider the importance of each text segment in the text set relative to the text to be processed, but also consider the similarity of the text segment to the determined part of the current summary content in each iteration. In this way, the more important text segments that are not highly redundant with the determined part of the current summary content can be dynamically screened out as target text segments, and the summary content can be continuously generated through a smaller number of target text segments. In this way, the number of text segments that need to be processed to generate the summary content can be reduced, and the efficiency of generating the summary content can be improved.

[0101] After determining the features of the text set, if the t-th time step is not the last time step among the multiple time steps, a partial summary of the summary content at the t-th time step is determined according to the features of the text set, and the summary content corresponding to the text set is continued to be generated through the t+1-th time step and other time steps.

[0102] The embodiment of the present application provides a summary determination method, for a text set including at least one text to be processed, in the process of determining the summary content corresponding to the text set, based on the importance parameter of the text fragment included in the text to be processed relative to the text to be processed, and the redundancy parameter relative to the determined part in the summary content, the summary weight of the text fragment is obtained together. Among them, the redundancy parameter is the similarity of the text fragment to the determined part of the current summary content when the summary content is determined this time. Through the redundancy parameter, it can be determined which text fragments in the text set have a high similarity with the determined part of the current summary content, and which text fragments have a low similarity. Since the text fragments with a high similarity based on the determined part of the current summary content are not very helpful for continuing to generate the summary content, therefore, according to the summary weight determined based on the importance parameter and the redundancy parameter, the text fragments included in the text set that are more important and have a low redundancy with the determined part of the current summary content can be screened out as target text fragments, and the summary content can be continued to be generated through a small number of target text fragments. In this way, the number of text fragments required to be processed to generate the summary content can be reduced, and the generation efficiency of the summary content can be improved.

[0103] In actual applications, users can enter a search string in the search interface provided by the terminal device, and search the user's intention in the form of a search string, and the search string includes at least one search term. For example, when a user wants to obtain relevant information about a certain event news, he can enter the search string in the search interface: a certain event. The terminal device can obtain a text set based on the search string, and the text set includes at least one text to be processed related to the search string. Among them, the search string is used to identify the user's search intention, and the text to be processed included in the text set is the search result based on the search string.

[0104] The text collection includes multiple text fragments, and the relevance of each text fragment to the user's intention may vary. For example, some text fragments are relevant to the user's intention, while some text fragments are irrelevant to the user's intention. In order to make the generated summary content more consistent with the user's intention, after screening out the text fragments that are helpful in generating the summary content, a third factor can be introduced: the relevance parameter.

[0105] The relevance parameter can be used to determine which text fragments in the text collection are more relevant to the user's intent and which text fragments are less relevant to the user's intent, thereby filtering out text fragments in the text collection that are helpful in generating summary content based on the relevance parameter.

[0106] The following is a method for obtaining a relevance parameter. First, the segment features of the text segment relative to the search string are determined based on the search string. Based on the segment features, the search string features of the search string, and the position information of the text segment in the text to be processed, the relevance parameter of the text segment relative to the search string is determined. For details, see formula (6):

[0107] r2 i =f(sl i ; ind1 i ; ind2 i ; l i ;simq i ]) (6)

[0108] Among them, r2 i represents the relevance parameter of the ith text segment in the text collection relative to the search string; sl i Indicates the length of the i-th text segment; idx1 i Indicates the absolute position of the i-th text segment in the text to be processed; idx2 i Indicates the relative position of the i-th text segment in the text to be processed; l i represents the segment feature of the i-th text segment; simq i Represents the cosine distance between the segment feature of the i-th text segment and the segment feature of the search string.

[0109] Specifically, the above parameters are described in detail below.

[0110] Among them, f( ) represents the parameter sl i ,idx1 i ,idx2 i , l i and simq i Get r2 i The embodiment of the present application does not specifically limit the method for obtaining the correlation parameter r2 i For example, r2 i It can also be obtained through the SVR model. Among them, the importance parameter r1 i and the correlation parameter r2 i The acquisition methods can be the same or different.

[0111] idx1 i with idx2 i is the position information of the text segment in the text to be processed, which respectively represents the absolute position and the relative position. Please refer to the above related instructions for details, which will not be repeated here.

[0112] It can be understood that when the correlation parameter is introduced, when the importance parameter is obtained by formula (1), the segment feature l in formula (1)i It is the text segment determined based on the search string relative to the segment features of the search string. Other similarities will not be repeated.

[0113] After the text set is determined based on the search string, the relevance parameters of each text fragment will not change with the process of determining the summary and are relatively fixed parameters. Therefore, in an embodiment of the present application, the determined relevance parameters can be recorded and directly called during the summary determination process without the need for repeated calculations.

[0114] The following describes how to obtain the segment feature li of the i-th text segment through the QBS-Encoder method.

[0115] When encoding text fragments, this encoding method takes into account the relevance of the text fragments to the user's intent. The fragment features of the text fragments obtained in this way will more prominently reflect the fragment features with a high degree of relevance to the search string. The generated fragment features strengthen the information of the text fragments related to the user's intent, and weaken the information of the text fragments irrelevant to the user's intent, thereby playing the role of selecting information in the fragment features based on the user's intent, so that it includes more information of the text fragments related to the user's intent.

[0116] The following uses a sentence as an example of a text segment to illustrate the QBS-Encoder method. The sentence includes multiple word segments, and the search string includes at least one search term. Based on each word segment vector and each search term vector, the sub-correlation parameter of the word segment relative to each search term in the search string can be first determined, and then the correlation parameter of the word segment relative to the search string can be determined based on the multiple sub-correlation parameters. This is described in detail below.

[0117] Determine the sub-relevance parameter of each word segmentation vector corresponding to each search word vector in the search string, see formula (7):

[0118] S ij =f s (h i ,u j ) (7)

[0119] Among them, S ij represents the sub-correlation parameter between the i-th word in the text segment and the j-th search term in the search string; the text segment includes I words, h i Represents the word segmentation vector of the i-th word segmentation, i∈I, i and I are both integers, I>1; the search string includes J search terms in total, u j Represents the search term vector of the j-th search term, j∈J, j and J are both integers, J>0.

[0120] f s (h i ,uj ) is the word segmentation vector h i and the search word vector u j Sub-correlation parameter calculation function, see formula (8):

[0121]

[0122] in, is a learnable parameter used to adjust the sub-correlation parameter calculation function; is the retrieval word vector u j With the word vector h i A composite function or a combination of functions.

[0123] After determining the sub-relevance parameter, the relevance parameter of each word segment relative to the search string can be obtained according to the sub-relevance parameter. For example, the obtained sub-relevance parameter can be subjected to a normalized exponential function softmax operation to obtain a relevance parameter, and the range of each relevance parameter is between (0, 1). See formula (9):

[0124] a' i =softmax(∑ J S ij ) (9)

[0125] Among them, a' i Represents the relevance parameter of the i-th word in the target text relative to the search string.

[0126] From the above, it can be seen that since the degree of relevance between each word segmentation and the search string may be different, the relevance parameters of each word segmentation may be different. When determining the fragment features of a text fragment, the fragment features of the text fragment are determined based on the relevance parameters corresponding to multiple word segmentations and the word segmentation vectors corresponding to multiple word segmentations.

[0127] For example, the representation of each word segmentation vector is updated by combining each word segmentation vector with its corresponding correlation parameter, and the segment feature of the text segment is obtained based on the updated word segmentation vector. The updated word segmentation vector can be obtained by referring to formula (10):

[0128] h' i =a' i ×h i (10)

[0129] Among them, h' i represents the updated i-th word segmentation vector; a' i represents the relevance parameter of the ith word relative to the search string; h i Represents the i-th word segmentation vector.

[0130] After obtaining all updated word segmentation vectors in the text segment, the updated word segmentation vector h' included in a text segment is i Splice and obtain the segment features l of the text segment i .

[0131] Thus, the segment feature l is obtained through the above formula (7)-formula (10): i The method is the QBS-Encoder method, which can be expressed as formula (11):

[0132] l k =QBSEncoder(d,q) (11)

[0133] Among them, d represents the text set and q represents the search string.

[0134] It can be understood that the segment feature l of the above-mentioned text segment is obtained i In the embodiment, the variables i and j used are only exemplary and have no special meaning, and may be the same as or different from the variables i and j in other parts of the embodiment of the present application.

[0135] After obtaining the relevance parameter, the summary weight of the text segment can be obtained by the importance parameter of the text segment relative to the text to be processed, the redundancy parameter of the text segment relative to the determined part of the summary content, and the relevance parameter of the text segment relative to the search string. Formula (3) can be transformed into Formula (12):

[0136] imp i =α1×r1 i +α2×r2 i -β×r3 i (12)

[0137] Among them, α1, α2 and β are weight adjustment parameters, for example, α1+α2+β=1.

[0138] Since the QBS-Encoder method is used for encoding, the decoding algorithm driven by user intent (Adapted Query-Driven decoder, AQD-Decoder) can be used for decoding, and the obtained text set features are input into the AQD-Decoder decoder, that is, the text set features are used as the input data of the AQD-Decoder decoder, and then decoded to generate the summary content corresponding to the text set.

[0139] Another method for obtaining importance parameters is described below. After obtaining the segment features of the text segment, the importance parameters of the text segment relative to the text set can be obtained based on the segment features. For example, the similarity between each text segment is first obtained based on the segment features, see formula (13):

[0140]

[0141] Among them, H ii' Represents the fragment feature l i and fragment features l i' The similarity between i' Represents the segment feature of the i'th text segment in the target text, i'∈I, i' is an integer.

[0142] When i≠i', the segment feature l i and fragment features l i' Not the same fragment feature, H ii’ for is the fragment feature l i and fragment features l i' The synthetic function or composite function of i = i', the fragment feature l i and fragment features l i' They are the same fragment features, and the similarity is 0.

[0143] Then, the similarities corresponding to all text fragments in the text set are summed and a softmax operation is performed to obtain the parameter b i , this parameter can reflect the importance of the text segment relative to the text set, see formula (14) for details:

[0144] b i =softmax(∑ I H ii' ) (14)

[0145] Among them, in the method of obtaining the importance parameter, due to b i It can reflect the importance of the text fragment relative to the text collection, so b i As a parameter r1 representing the importance of the i-th text segment in the text collection relative to the text to be processed i .

[0146] In addition, in formula (1), in addition to determining simd by calculating the cosine distance i In addition, the embodiment of the present application also provides another method for determining simd based on formula (1): i An optional method, in which the parameter b obtained by combining formula (14) i (In the absence of bi As r1 i ), and the l determined by formula (11) i Generate text features of the text to be processed, see formula (15):

[0147] F'=∑ I (l i ×b i ) (15)

[0148] Among them, F' represents the text feature corresponding to a text to be processed; l i Represents the segment features of the i-th text segment.

[0149] Therefore, we can use the segment feature l of the i-th text segment i The cosine distance between the text feature F' of the text to be processed is obtained by simd i .

[0150] The summary determination method provided in the embodiment of the present application can be applied to search engines, intelligent customer service, personalized browsing and other scenarios. For example, in the search engine scenario, by dynamically generating summary content that meets the user's intention, it can help users quickly obtain content of interest from a large amount of text; in the intelligent customer service scenario, it provides a solution to the user's query content in the form of dynamically generated summary content; in the personalized browsing scenario, it summarizes the summary content that contains the user's intention so that the user can quickly browse the corresponding content, etc.

[0151] Next, the summary determination method provided by the embodiment of the present application is described by taking the search engine scenario as an example. Figure 3 , this figure is a schematic diagram of an application scenario of a summary determination method provided in an embodiment of the present application, which shows the process of generating summary content in the form of pseudo code.

[0152] like Figure 3 As shown, multiple to-be-processed texts SDS, a text set D and a search string q are taken as inputs, and the summary determination method provided in the embodiment of the present application is adopted to obtain the summary content corresponding to the text set D. The following is a detailed description.

[0153] The user enters a search string q in the search interface provided by the terminal device. The search string q is used to identify the user's search intention. The terminal device obtains a text set D according to the search string q, wherein the text set D includes multiple to-be-processed texts SDS, each of which includes multiple text segments, and the text set includes I text segments in total. Each text segment in the text set D can be represented as s i ,s i ∈D, i∈I, i and I are both integers, I>1.

[0154] As can be seen from the above, the generation of summary content is gradually completed based on a sequential process. Lines 1 to 7 of the pseudo code are the preparations before entering the iterative process.

[0155] Line 1: Use the SDS of the text to be processed to train the AQD-Decoder model so that the attention weights corresponding to the text segments can be obtained according to the AQD-Decoder model, and the target text segments determined at each time step can be decoded.

[0156] Line 2: The SVR model uses formula (1) to determine the importance parameter of the text segment relative to the text to be processed based on the segment features, the text features of the text to be processed, and the position information of the text segment in the text to be processed.

[0157] Line 3: The SVR model is used to determine the relevance parameter of each text segment in the text set relative to the search string based on the segment features, the search string features, and the position information of the text segment in the text to be processed using formula (6).

[0158] Line 4: The summary weight of the text segment is determined according to the importance parameter, the relevance parameter, and the redundancy parameter through formula (12).

[0159] Line 5: However, since the summary content has not been obtained at this time, the redundancy parameter is 0, and the summary weight imp of the i-th text segment in the text set D is i The method of obtaining can be referred to formula (16):

[0160] imp i =α1×r1 i +α2×r2 i (16)

[0161] Line 6: The summary content corresponding to the text set D is determined through multiple time steps. One time step corresponds to one iteration to generate a partial summary, thereby establishing an empty set S = {} of summary content to store the partial summary generated at each time step.

[0162] Line 7: At this point, the length of the determined portion of the summary content in the set S={} is 0.

[0163] After completing the preparation work, the summary content corresponding to the text set D is obtained through multiple iterations. Pseudocode lines 8 to 17 represent the iterative process. The following description is based on the partial summary determined at each time step as the vocabulary, and the length of the summary content is the number of words included.

[0164] In order to filter out a small number of target text fragments from the text fragments included in the text set D, so as to obtain the summary content corresponding to the text set D. When decoding using the aforementioned AQD-Decoder, an improved maximum boundary correlation model (Adapted Maximal Marginal Relevance Model, AMMR) is used, hereinafter referred to as the AQD-AMMR model. When decoding using the AQD-AMMR model, each time step will obtain the importance parameter of the text fragment relative to the text to be processed, the redundancy parameter of the text fragment relative to the determined part in the summary content, and the relevance parameter of the text fragment relative to the search string, so as to filter out the target text fragment.

[0165] Line 8: The length of the summary content is limited. In this embodiment, it is set to L. max , and gradually obtain the summary content corresponding to the text set D through multiple time steps until the length of the obtained summary content meets the set length. <L max , do the following:

[0166] Line 9: Get the summary weight of each text segment, sort the I text segments included in the text set D, select the top K text segments with the largest summary weights as target text segments, and form the target text segment set SK.

[0167] Line 10: Use the AQD-Decoder model to obtain the attention weight corresponding to the target text segment, and according to the summary weight of the target text segment at the tth time step Adjust the attention weight corresponding to the target text segment at the tth time step when generating the text set features corresponding to the text set And obtain the text set features corresponding to the text set D See formula (4) and formula (5).

[0168] Line 11: Set the text collection features Input to the AQD-Decoder model for decoding, and obtain the partial summary S corresponding to the tth time step t .

[0169] Line 12: The partial summary S is obtained t Add to the set S of summary content.

[0170] Line 13: The SVR model is used to determine the importance parameter of the text segment relative to the text to be processed based on the segment features, the text features of the text to be processed, and the position information of the text segment in the text to be processed using formula (1).

[0171] Line 14: The SVR model is used to determine the relevance parameter of each text segment in the text set relative to the search string based on the segment features, the search string features, and the position information of the text segment in the text to be processed using formula (6).

[0172] It should be noted that, since the importance parameters and relevance parameters are relatively fixed parameters, the importance parameters and relevance parameters obtained in the 2nd and 3rd rows respectively can be recorded and directly called when executing the operations in the 12th and 13th rows, without the need for repeated calculations.

[0173] Line 15: A part of the vocabulary of the summary content can be generated at each time step. Correspondingly, at each time step, the redundancy parameter corresponding to the text segment will change due to the change of the determined part of the current summary content. The redundancy parameter relative to the determined part of the summary content can be obtained by formula (2).

[0174] Line 16: Thus, the summary weight of the text fragment can be obtained through formula (12) by using the importance parameter of the text fragment relative to the text to be processed, the redundancy parameter of the text fragment relative to the determined part in the summary content, and the relevance parameter of the text fragment relative to the search string.

[0175] Line 17: Continue the above steps until the length of the summary content s ≥ L max , get the set S at this time.

[0176] The probability distribution of the words included in the set S is calculated, see formula (17):

[0177]

[0178] Among them, V1, V2, s t , b1 and b2 are all learnable parameters, for example, parameters of a deep learning network; Represents the text set features corresponding to the text set; s t is the state of the decoder at time step t.

[0179] Specifically, the state s t Used to generate the partial summary S corresponding to the t-th time step t For example, the decoding process can be understood as a unidirectional recurrent neural network (RNN), according to the state s at the t-1 time step t-1 Generate the state of the tth time step, see formula (18):

[0180] s t =f(s <t-1> ,y <t-1>) (18)

[0181] Among them, y <t-1>represents the vocabulary generated by decoding at the t-1th time step, s <t-1>represents the decoder state at the t-1th time step.

[0182] In view of the summary determination method provided in the above embodiment, the embodiment of the present application further provides a summary determination device.

[0183] See also Figure 4 , which is a schematic diagram of the structure of a summary determination device provided in an embodiment of the present application. Figure 4 As shown, the summary determination device 400 includes: an acquisition unit 401, a first determination unit 402, a second determination unit 403 and a generation unit 404;

[0184] The acquisition unit 401 is used to acquire a text set including at least one text to be processed, wherein the text to be processed includes multiple text segments;

[0185] The first determining unit 402 is used to obtain the summary weight of the text segment by using the importance parameter of the text segment relative to the text to be processed and the redundancy parameter of the text segment relative to the determined part of the summary content in the process of determining the summary content corresponding to the text set;

[0186] The second determining unit 403 is used to determine target text segments from the text segments included in the text set based on the summary weight, wherein the number of the target text segments is less than the number of the text segments in the text set;

[0187] The generating unit 404 is configured to generate the summary content corresponding to the text set according to the target text segment.

[0188] As a possible implementation manner, the summary content corresponding to the text set is determined through multiple time steps, and for the i-th time step among the multiple time steps, the device 400 is used to:

[0189] The partial summary determined in the first i-1 time steps is used as the determined part of the summary content, and the redundancy parameter of the text segment in the i-th time step is determined.

[0190] As a possible implementation manner, the second determining unit 403 is configured to:

[0191] Determining a target text segment for the i-th time step from the text segments included in the text set according to the summary weight;

[0192] The generating unit 404 is used for:

[0193] Generate a text set feature corresponding to the text set through the target text segment at the i-th time step;

[0194] A partial summary of the summary content at the i-th time step is determined according to the text set features.

[0195] As a possible implementation manner, the generating unit 404 is configured to:

[0196] According to the summary weight of the target text segment at the i-th time step, adjusting the attention weight corresponding to the target text segment at the i-th time step when generating the text set feature corresponding to the text set;

[0197] The text set features corresponding to the text set are generated by using the adjusted attention weights and the segment features of the target text segment at the i-th time step.

[0198] As a possible implementation manner, the text set is determined by a search string, and the first determining unit 402 is used to:

[0199] The summary weight of the text segment is obtained by the importance parameter of the text segment relative to the text to be processed, the redundancy parameter of the text segment relative to the determined part in the summary content, and the relevance parameter of the text segment relative to the search string.

[0200] As a possible implementation manner, the device 400 is used to:

[0201] Determining, based on the search string, a segment feature of the text segment relative to the search string;

[0202] Determining a relevance parameter of the text segment relative to the search string according to the segment feature, the search string feature of the search string, and the position information of the text segment in the text to be processed;

[0203] An importance parameter of the text segment relative to the text to be processed is determined according to the segment feature, the text feature of the text to be processed, and the position information of the text segment in the text to be processed.

[0204] As a possible implementation manner, the text set includes a plurality of texts to be processed, and the second determining unit 403 is configured to:

[0205] The top K text segments with the largest summary weights are determined from the text segments respectively included in the multiple texts to be processed as the target text segments.

[0206] The summary determination device provided in the above embodiment, for a text set including at least one text to be processed, in the process of determining the summary content corresponding to the text set, based on the importance parameter of the text segment included in the text to be processed relative to the text to be processed, and the redundancy parameter relative to the determined part in the summary content, jointly obtains the summary weight of the text segment. Among them, the redundancy parameter is the similarity of the text segment to the determined part of the current summary content when the summary content is determined this time. Through the redundancy parameter, it can be determined which text segments in the text set have a high similarity with the determined part of the current summary content, and which text segments have a low similarity. Since the text segments with a high similarity based on the determined part of the current summary content are not very helpful for continuing to generate the summary content, therefore, according to the summary weight determined based on the importance parameter and the redundancy parameter, the more important text segments and the text segments with low redundancy with the determined part of the current summary content can be screened from the text segments included in the text set as the target text segments, and the summary content is continued to be generated through a small number of target text segments. In this way, the number of text segments required to be processed to generate the summary content can be reduced, and the generation efficiency of the summary content is improved.

[0207] The embodiment of the present application also provides a computer device. The computer device provided by the embodiment of the present application will be introduced from the perspective of hardware instantiation.

[0208] See also Figure 5 , Figure 5 14 is a schematic diagram of a server structure provided in an embodiment of the present application. The server 1400 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1422 (for example, one or more processors) and memory 1432, and one or more storage media 1430 (for example, one or more mass storage devices) storing application programs 1442 or data 1444. Among them, the memory 1432 and the storage medium 1430 may be short-term storage or permanent storage. The program stored in the storage medium 1430 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1422 may be configured to communicate with the storage medium 1430 to execute a series of instruction operations in the storage medium 1430 on the server 1400.

[0209] The server 1400 may also include one or more power supplies 1426, one or more wired or wireless network interfaces 1450, one or more input and output interfaces 1458, and / or one or more operating systems 1441, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0210] The steps performed by the server in the above embodiment can be based on the Figure 5 The server structure shown.

[0211] The CPU 1422 is used to perform the following steps:

[0212] Acquire a text set including at least one text to be processed, wherein the text to be processed includes a plurality of text segments;

[0213] In the process of determining the summary content corresponding to the text set, the summary weight of the text segment is obtained by using the importance parameter of the text segment relative to the text to be processed and the redundancy parameter of the text segment relative to the determined part of the summary content;

[0214] Determining target text segments from the text segments included in the text collection based on the summary weights, wherein the number of the target text segments is less than the number of the text segments in the text collection;

[0215] The summary content corresponding to the text collection is generated according to the target text segment.

[0216] Optionally, the CPU 1422 may also execute method steps of any specific implementation of the abstract determination method in the embodiments of the present application.

[0217] With respect to the summary determination method described above, an embodiment of the present application further provides a terminal device for summary determination, so that the summary determination method described above can be implemented and applied in practice.

[0218] See also Figure 6 , Figure 6 A schematic diagram of the structure of a terminal device provided in an embodiment of the present application. For the sake of convenience, only the parts related to the embodiment of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present application. The terminal device can be any terminal device including a mobile phone, a tablet computer, a personal digital assistant (PDA), etc., taking the terminal device as a mobile phone as an example:

[0219] Figure 6 FIG. 1 is a block diagram showing a partial structure of a mobile phone related to a terminal device provided in an embodiment of the present application. Figure 6 The mobile phone includes: a radio frequency (RF) circuit 1510, a memory 1520, an input unit 1530, a display unit 1540, a sensor 1550, an audio circuit 1560, a wireless fidelity (WiFi) module 1570, a processor 1580, and a power supply 1590. Those skilled in the art will understand that Figure 6 The mobile phone structure shown in the figure does not constitute a limitation on the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0220] Combine the following Figure 6 A detailed introduction to the various components of the mobile phone:

[0221] The RF circuit 1510 can be used for receiving and sending signals during the process of sending and receiving information or making calls. In particular, after receiving the downlink information of the base station, it is sent to the processor 1580 for processing; in addition, the designed uplink data is sent to the base station. Usually, the RF circuit 1510 includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (Low Noise Amplifier, referred to as LNA), a duplexer, etc. In addition, the RF circuit 1510 can also communicate with the network and other devices through wireless communication. The above wireless communication can use any communication standard or protocol, including but not limited to the Global System of Mobile communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0222] The memory 1520 can be used to store software programs and modules. The processor 1580 implements various functional applications and data processing of the mobile phone by running the software programs and modules stored in the memory 1520. The memory 1520 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 1520 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0223] The input unit 1530 can be used to receive input digital or character information, and to generate key signal input related to the user settings and function control of the mobile phone. Specifically, the input unit 1530 may include a touch panel 1531 and other input devices 1532. The touch panel 1531, also known as a touch screen, can collect the user's touch operation on or near it (such as the user's operation on the touch panel 1531 or near the touch panel 1531 using any suitable object or accessory such as a finger, stylus, etc.), and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel 1531 may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into the touch point coordinates, and then sends it to the processor 1580, and can receive and execute the command sent by the processor 1580. In addition, the touch panel 1531 can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic waves. In addition to the touch panel 1531, the input unit 1530 may also include other input devices 1532. Specifically, other input devices 1532 may include but are not limited to one or more of a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and the like.

[0224] The display unit 1540 may be used to display information input by the user or information provided to the user and various menus of the mobile phone. The display unit 1540 may include a display panel 1541. Optionally, the display panel 1541 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. Further, the touch panel 1531 may cover the display panel 1541. When the touch panel 1531 detects a touch operation on or near it, it is transmitted to the processor 1580 to determine the type of touch event. Subsequently, the processor 1580 provides a corresponding visual output on the display panel 1541 according to the type of touch event. Although in Figure 6 In the embodiment, the touch panel 1531 and the display panel 1541 are used as two independent components to realize the input and output functions of the mobile phone, but in some embodiments, the touch panel 1531 and the display panel 1541 can be integrated to realize the input and output functions of the mobile phone.

[0225] The mobile phone may also include at least one sensor 1550, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 1541 according to the brightness of the ambient light, and the proximity sensor may turn off the display panel 1541 and / or the backlight when the mobile phone is moved to the ear. As a type of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that identify the posture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors that can be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., they will not be repeated here.

[0226] The audio circuit 1560, the speaker 1561, and the microphone 1562 can provide an audio interface between the user and the mobile phone. The audio circuit 1560 can transmit the received audio data to the speaker 1561 after converting the received audio data into an electrical signal, which is converted into a sound signal for output; on the other hand, the microphone 1562 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1560 and converted into audio data, and then the audio data is output to the processor 1580 for processing, and then sent to another mobile phone through the RF circuit 1510, or the audio data is output to the memory 1520 for further processing.

[0227] WiFi is a short-range wireless transmission technology. The mobile phone can help users send and receive emails, browse web pages and access streaming media through the WiFi module 1570. It provides users with wireless broadband Internet access. Figure 6 A WiFi module 1570 is shown, but it is understandable that it is not an essential component of the mobile phone and can be omitted as needed without changing the essence of the invention.

[0228] The processor 1580 is the control center of the mobile phone. It uses various interfaces and lines to connect various parts of the entire mobile phone. It executes various functions of the mobile phone and processes data by running or executing software programs and / or modules stored in the memory 1520 and calling data stored in the memory 1520. Optionally, the processor 1580 may include one or more processing units; preferably, the processor 1580 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 1580.

[0229] The mobile phone also includes a power supply 1590 (such as a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the processor 1580 through a power management system, so that the power management system can manage functions such as charging, discharging, and power consumption.

[0230] Although not shown, the mobile phone may also include a camera, a Bluetooth module, etc., which will not be described in detail here.

[0231] In the embodiment of the present application, the memory 1520 included in the mobile phone can store program codes and transmit the program codes to the processor.

[0232] The processor 1580 included in the mobile phone can execute the digest determination method provided in the above embodiment according to the instructions in the program code.

[0233] The embodiment of the present application further provides a computer-readable storage medium for storing a computer program, wherein the computer program is used to execute the summary determination method provided in the above embodiment.

[0234] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the summary determination method provided in various optional implementations of the above aspects.

[0235] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the above-mentioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the above-mentioned storage medium can be at least one of the following media: read-only memory (English: read-only memory, abbreviated: ROM), RAM, magnetic disk or optical disk, etc. Various media that can store program codes.

[0236] It should be noted that each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0237] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims. < / t-1>

Claims

1. A method for determining a summary, characterized in that: The method is applied to a scenario where summary content is continuously generated, and the method includes: Acquiring a text set including at least one text to be processed through a search string, wherein the text to be processed includes a plurality of text segments, and the search string is used to identify the search intention of the user; In the process of determining the summary content corresponding to the text set, the summary weight of the text fragment is obtained by weighting the importance parameter of the text fragment relative to the text to be processed, the redundancy parameter of the text fragment relative to the determined part of the summary content, and the relevance parameter of the text fragment relative to the search string; the importance parameter is used to characterize the relevance degree of the text fragment to the text to be processed in which it is located, and the redundancy parameter is the similarity degree of the text fragment when determining the summary content this time relative to the determined part of the current summary content; Determining a target text segment for the t-th time step from the text segments included in the text set according to the summary weight, wherein the number of the target text segments is less than the number of the text segments in the text set; According to the summary weight of the target text segment at the t-th time step, adjusting the attention weight corresponding to the target text segment at the t-th time step when generating the text set feature corresponding to the text set, generating the text set feature corresponding to the text set by using the adjusted attention weight and the segment feature of the target text segment at the t-th time step, and decoding according to the text set feature to obtain the partial vocabulary of the summary content at the t-th time step; The summary content corresponding to the text set is determined through multiple time steps, and for the t-th time step among the multiple time steps, the part of the vocabulary determined in the previous t-1 time step is used as the determined part of the summary content, and the redundancy parameter of the text segment in the t-th time step is determined; The process of determining the importance parameter of the text segment relative to the text to be processed includes: Determining, based on the search string, a segment feature of the text segment relative to the search string; Inputting the segment feature, the search string feature of the search string, and the position information of the text segment in the text to be processed into a first model to determine a relevance parameter of the text segment relative to the search string; The segment feature, the text feature of the text to be processed and the position information of the text segment in the text to be processed are input into the second model to determine the importance parameter of the text segment relative to the text to be processed.

2. The method according to claim 1, characterized in that The text set includes a plurality of to-be-processed texts, and determining a target text segment for the t-th time step from the text segments included in the text set according to the summary weight includes: The top K text segments with the largest summary weights are determined from the text segments respectively included in the multiple texts to be processed as the target text segments for the t-th time step.

3. A summary determination device, characterized in that: The device is suitable for a scenario where summary content is continuously generated, and comprises: an acquisition unit, a first determination unit, a second determination unit and a generation unit; The acquisition unit is used to acquire a text set including at least one text to be processed through a search string, wherein the text to be processed includes a plurality of text segments, and the search string is used to identify the search intention of the user; The first determination unit is used to obtain the summary weight of the text segment by weighting the importance parameter of the text segment relative to the text to be processed, the redundancy parameter of the text segment relative to the determined part of the summary content, and the relevance parameter of the text segment relative to the search string in the process of determining the summary content corresponding to the text set; the importance parameter is used to characterize the relevance degree of the text segment to the text to be processed in which it is located, and the redundancy parameter is the similarity degree of the text segment when determining the summary content this time relative to the determined part of the current summary content; The second determining unit is used to determine a target text segment for the t-th time step from the text segments included in the text set according to the summary weight, wherein the number of the target text segments is less than the number of the text segments in the text set; The generating unit is used to adjust the attention weight corresponding to the target text segment at the t-th time step when generating the text set feature corresponding to the text set according to the summary weight of the target text segment at the t-th time step, generate the text set feature corresponding to the text set by using the adjusted attention weight and the segment feature of the target text segment at the t-th time step, and obtain the partial vocabulary of the summary content at the t-th time step by decoding according to the text set feature; The summary content corresponding to the text set is determined through multiple time steps. For the i-th time step among the multiple time steps, the device is used to: use the part of words determined in the previous t-1 time steps as the determined part of the summary content, and determine the redundancy parameter of the text segment in the t-th time step; The process of determining the importance parameter of the text segment relative to the text to be processed includes: Determining, based on the search string, a segment feature of the text segment relative to the search string; Inputting the segment feature, the search string feature of the search string, and the position information of the text segment in the text to be processed into a first model to determine a relevance parameter of the text segment relative to the search string; The segment feature, the text feature of the text to be processed and the position information of the text segment in the text to be processed are input into the second model to determine the importance parameter of the text segment relative to the text to be processed.

4. The device according to claim 3, characterized in that The text set includes a plurality of texts to be processed, and the second determining unit is used to: The top K text segments with the largest summary weights are determined from the text segments respectively included in the multiple texts to be processed as the target text segments for the t-th time step.

5. A computer device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the method according to any one of claims 1-2 according to the instructions in the program code.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 2.

7. A computer program product, characterized in that The computer program product comprises instructions, which, when executed on a computer device, cause the computer device to perform the method according to any one of claims 1 to 2.