Conversation summary generation method and device, electronic equipment and storage medium

By finding similar dialogue texts and their summary from the dialogue knowledge base based on dialogue length and semantics in complex business scenarios, generating model prompt information, and using a big model to perform dialogue summary, the accuracy and high cost of dialogue summary in the existing technology are solved. The problem of the problem has achieved an efficient and accurate dialogue summary.

CN119089909APending Publication Date: 2024-12-06SHENZHEN YISHIHUOLALA TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411190708.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

In the complex business scenarios, it is difficult to obtain high-quality training data in the existing technology, which makes it difficult to guarantee the accuracy of large-scale dialogue summary and is costly.

Method used

By obtaining the conversation to be summarized, searching for similar conversation texts and their summary from the pre-constructed conversation knowledge base based on the conversation length and semantics, generating model prompt information, and using a big model to summarize, reducing the cost of conversation summary.

Benefits of technology

It improves the accuracy of dialogue summary, reduces dependence on high-quality training data, reduces costs, and realizes effective dialogue summary in complex business scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089909A_ABST
    Figure CN119089909A_ABST
Patent Text Reader

Abstract

The invention discloses a dialogue summary generation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a current to-be-summarized dialogue; based on the dialogue length and the dialogue semantics, searching a plurality of similar dialogue texts of the current to-be-summarized dialogue and dialogue summaries corresponding to the similar dialogue texts from a pre-constructed dialogue knowledge base; generating model prompt information corresponding to the current to-be-summarized dialogue by utilizing the current to-be-summarized dialogue, the searched similar dialogue texts and the corresponding dialogue summaries; and inputting the model prompt information corresponding to the current to-be-summarized dialogue into a large model, and generating a dialogue summary corresponding to the current to-be-summarized dialogue through the large model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of text processing technology, and in particular to a conversation summary generation method and device, electronic device, and storage medium. Background Art

[0002] Currently, some companies need to invite multiple people to participate in their business. In order to facilitate the follow-up of the invitation and subsequent tracing of the invitees, the conversation information between the inviter and the invitee is currently recorded through the invitation center.

[0003] Since the original conversation between the inviter and the invitee is usually long and contains a lot of unnecessary information, which is not convenient for direct use, it needs to be summarized. The way of summarizing and recording by personnel is not only less complete, but also each person has different annotation habits, which makes it difficult to identify the recorded information and inconvenient to use. Therefore, the conversation information can be summarized through a large model. Specifically, it is necessary to enhance the business data first, that is, to increase the conversation information, and use the enhanced sample data to build a fine-tuning data set, and fine-tune the large model, so that the fine-tuned large model can be used to summarize the conversation text.

[0004] However, fine-tuning a large model requires a large amount of high-quality training data to train the model, which makes the annotation of training data in complex business scenarios not only costly but also difficult. Therefore, it is difficult to obtain a large amount of high-quality training data, and thus it is impossible to guarantee the accuracy of the text summarized by the large model. Summary of the invention

[0005] Based on the above-mentioned deficiencies of the prior art, the present application provides a conversation summary generation method and device, electronic device, and storage medium to solve the problem that the prior art is high in cost and cannot ensure the accuracy of the summary text.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of the present application provides a method for generating a conversation summary, comprising:

[0008] Get the current conversation to be summarized;

[0009] Based on the conversation length and conversation semantics, searching from a pre-built conversation knowledge base, a plurality of similar conversation texts of the current conversation to be summarized and conversation summaries corresponding to each of the similar conversation texts;

[0010] Generate model prompt information corresponding to the current conversation to be summarized by using the current conversation to be summarized, each of the similar conversation texts found and their corresponding conversation summaries;

[0011] The model prompt information corresponding to the current dialogue to be summarized is input into the large model, and a dialogue summary corresponding to the current dialogue to be summarized is generated by the large model.

[0012] Optionally, in the above-mentioned conversation summary generation method, the following is further included:

[0013] The conversation knowledge base is constructed using a plurality of example data; wherein one of the example data at least includes a conversation text example and a conversation summary corresponding to the conversation text example;

[0014] Performing distribution statistics on the lengths of each of the dialogue text examples in the dialogue knowledge base;

[0015] According to the amount of information in the dialogue text, the dialogue knowledge base is bucketed according to the text length to obtain multiple dialogue knowledge base length intervals.

[0016] Optionally, in the above-mentioned method for generating a conversation summary, searching, based on the conversation length and the conversation semantics, from a pre-built conversation knowledge base, a plurality of similar conversation texts of the current conversation to be summarized and the conversation summaries corresponding to the respective similar conversation texts include:

[0017] Retrieving from the conversation knowledge base a conversation knowledge base length interval to which the current conversation to be summarized belongs;

[0018] Calculate the similarity between the current dialogue to be summarized and each of the dialogue text examples within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs, using the text vector of the current dialogue to be summarized and the text vectors of each of the dialogue text examples within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs;

[0019] The first K dialogue text examples with the greatest similarity to the current dialogue to be summarized and their corresponding dialogue summaries are obtained to obtain each of the similar dialogue texts of the dialogue to be summarized and the dialogue summaries corresponding to each of the similar dialogue texts.

[0020] Optionally, in the above-mentioned conversation summary generation method, the step of using the current conversation to be summarized, each of the similar conversation texts found and their corresponding conversation summaries to generate model prompt information corresponding to the current conversation to be summarized includes:

[0021] Each of the similar dialogue texts and its corresponding dialogue summary is added to the model prompt information template as a reference example, and the current dialogue to be summarized is added to the model prompt information template as a target to be summarized, so as to obtain the model prompt information corresponding to the current dialogue to be summarized.

[0022] Optionally, in the above-mentioned method for generating a conversation summary, after inputting the model prompt information corresponding to the current conversation to be summarized into the big model and generating the conversation summary corresponding to the current conversation to be summarized by using the big model, the method further includes:

[0023] Determining the intent tags of each of the similar dialogue texts;

[0024] According to the majority voting principle, the intention tag with the largest number among the intention tags of the similar conversation texts is used as the intention tag of the conversation summary corresponding to the current conversation to be summarized.

[0025] Optionally, in the above-mentioned method for generating a conversation summary, determining the intention label of each of the similar conversation texts includes:

[0026] From the conversation knowledge base, find out the intention tags of each of the similar conversation texts; wherein the intention tags of the similar conversation texts are pre-matched based on the reason tags in the conversation summaries corresponding to the similar conversation texts, and stored in the conversation knowledge base.

[0027] Optionally, in the above-mentioned method for generating a conversation summary, after the intention tag with the largest number among the intention tags of each of the similar conversation texts is used as the intention tag of the conversation summary corresponding to the current conversation to be summarized according to the majority voting principle, the method further includes:

[0028] The conversation summary and its intention label corresponding to the current conversation to be summarized are output as the intelligent summary of the current conversation to be summarized.

[0029] A second aspect of the present application provides a conversation summary generating device, comprising:

[0030] A dialogue acquisition unit, used to acquire the current dialogue to be summarized;

[0031] A conversation search unit, configured to search, based on the conversation length and conversation semantics, from a pre-built conversation knowledge base for a plurality of similar conversation texts of the current conversation to be summarized and conversation summaries corresponding to the respective similar conversation texts;

[0032] A generating unit, configured to generate model prompt information corresponding to the current conversation to be summarized by using the current conversation to be summarized, each of the similar conversation texts found and their corresponding conversation summaries;

[0033] The summarizing unit is used to input the model prompt information corresponding to the current dialogue to be summarized into the large model, and generate a dialogue summary corresponding to the current dialogue to be summarized through the large model.

[0034] Optionally, in the above-mentioned conversation summary generating device, it also includes:

[0035] A construction unit, configured to construct the dialogue knowledge base using a plurality of example data; wherein one of the example data at least includes a dialogue text example and a dialogue summary corresponding to the dialogue text example;

[0036] A statistical unit, used for collecting distribution statistics of the lengths of the respective dialogue text examples in the dialogue knowledge base;

[0037] The division unit is used to divide the dialogue knowledge base into buckets according to the text length according to the amount of information in the dialogue text, so as to obtain multiple dialogue knowledge base length intervals.

[0038] Optionally, in the above-mentioned conversation summary generating device, the conversation searching unit includes:

[0039] A retrieval unit, used to retrieve from the dialogue knowledge base the dialogue knowledge base length interval to which the current dialogue to be summarized belongs;

[0040] a calculation unit, configured to calculate the similarity between the current dialogue to be summarized and each of the dialogue text examples within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs, by using the text vector of the current dialogue to be summarized and the text vectors of each of the dialogue text examples within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs;

[0041] The extraction unit is used to obtain the first K dialogue text examples with the greatest similarity to the current dialogue to be summarized and their corresponding dialogue summaries, and obtain each of the similar dialogue texts of the dialogue to be summarized and the dialogue summary corresponding to each of the similar dialogue texts.

[0042] Optionally, in the above-mentioned conversation summary generating device, the generating unit includes:

[0043] A generation subunit is used to add each of the similar dialogue texts and its corresponding dialogue summary as a reference example to the model prompt information template, and add the current dialogue to be summarized as a target to be summarized to the model prompt information template to obtain the model prompt information corresponding to the current dialogue to be summarized.

[0044] Optionally, in the above-mentioned conversation summary generating device, it also includes:

[0045] A label determination unit, used to determine the intention label of each of the similar dialogue texts;

[0046] The voting unit is used to, according to the majority voting principle, use the intention tag with the largest number among the intention tags of each of the similar dialogue texts as the intention tag of the dialogue summary corresponding to the current dialogue to be summarized.

[0047] Optionally, in the above-mentioned conversation summary generating device, the label determining unit includes:

[0048] A label search unit is used to search for the intention label of each of the similar dialogue texts from the dialogue knowledge base; wherein the intention label of the similar dialogue text is pre-matched based on the reason label in the dialogue summary corresponding to the similar dialogue text, and stored in the dialogue knowledge base.

[0049] Optionally, in the above-mentioned conversation summary generating device, it also includes:

[0050] The output unit is used to output the conversation summary and its intention label corresponding to the current conversation to be summarized as the intelligent summary of the current conversation to be summarized.

[0051] A third aspect of the present application provides an electronic device, including:

[0052] Memory and processor;

[0053] Wherein, the memory is used to store programs;

[0054] The processor is used to execute the program, and when the program is executed, it is specifically used to implement the conversation summary generation method as described in any one of the above items.

[0055] A fourth aspect of the present application provides a computer storage medium for storing a computer program, which, when executed by a processor, is used to implement a conversation summary generation method as described in any one of the above.

[0056] The present application provides a method for generating a dialogue summary, which obtains the current dialogue to be summarized, and based on the dialogue length and dialogue semantics, finds out multiple similar dialogue texts of the current dialogue to be summarized and the dialogue summaries corresponding to each of the similar dialogue texts from a pre-constructed dialogue knowledge base, so that by finding dialogue texts similar to the current dialogue to be summarized in length and semantics, the information contained and the dialogue texts similar in semantics to the current dialogue to be summarized can be obtained, so that the dialogue summary of the current dialogue to be summarized can be made with reference to the similar dialogue texts and their corresponding dialogue summaries. Then, the model prompt information corresponding to the current dialogue to be summarized is generated by using the current dialogue to be summarized, each of the similar dialogue texts found out and their corresponding dialogue summaries, so that the dynamic model prompt information can be constructed to more accurately summarize the current dialogue to be summarized. Finally, the model prompt information corresponding to the current dialogue to be summarized is input into the large model, and the dialogue summary corresponding to the current dialogue to be summarized is generated by the large model, that is, under the guidance of the model prompt information, the large model summarizes the current dialogue to be summarized with reference to the similar dialogue texts and their corresponding dialogue summaries, thereby effectively ensuring the accuracy of the dialogue summary, not requiring a large amount of high-quality data, and reducing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] 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 embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0058] Figure 1 A flowchart of a method for generating a conversation summary provided in an embodiment of the present application;

[0059] Figure 2 A flowchart of a method for constructing a conversation knowledge base provided in an embodiment of the present application;

[0060] Figure 3 A flowchart of a method for searching similar conversation texts and their corresponding conversation summaries provided in an embodiment of the present application;

[0061] Figure 4 A flowchart of a method for analyzing an intention to receive an invitation provided in an embodiment of the present application;

[0062] Figure 5 A schematic diagram of the architecture of a conversation summary generating device provided in an embodiment of the present application;

[0063] Figure 6 A schematic diagram of the architecture of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0065] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0066] The embodiment of the present application provides a method for generating a conversation summary, such as Figure 1 As shown, the specific steps include:

[0067] S101. Obtain the current dialogue to be summarized.

[0068] The current dialogue to be summarized is the dialogue text that currently needs to be summarized.

[0069] S102: Based on the conversation length and conversation semantics, multiple similar conversation texts of the current conversation to be summarized and conversation summaries corresponding to the respective similar conversation texts are searched from a pre-built conversation knowledge base.

[0070] It should be noted that in the embodiment of the present application, by generating model prompt information for summarizing the conversation, the large model is guided to use its own speech comprehension and generation capabilities to complete the conversation summary. Considering the diverse conversation scenarios and responsible conversation scenarios, in the embodiment of the present application, the method of setting a fixed model prompt information prompt based on experience is not adopted, but multiple similar conversation texts and their corresponding conversation summaries of the current conversation to be summarized are found, and a dynamic model prompt information prompt is constructed, so that the large model can be guided to generate the conversation summary of the current conversation to be summarized through similar conversation texts and their corresponding conversation summaries, thereby ensuring the accuracy of the generated conversation summary.

[0071] Therefore, in the embodiment of the present application, the conversation texts in multiple scenarios are collected, and the corresponding conversation summaries are constructed to construct a conversation knowledge base. In addition, in the subsequent use process, the currently generated conversation summary can be calibrated and added to the conversation knowledge base, so as to continuously increase the amount of data in the conversation knowledge base, so as to facilitate more accurate conversation summaries of various conversation texts.

[0072] Since the amount of information protected and the information distribution location of conversation texts of different lengths are usually inconsistent, and the amount of data that needs to be reduced is also different, and the simplified texts of the same or similar lengths usually have similar amounts of information and similar information distribution. Therefore, in the embodiment of the present application, when searching for similar conversation texts of the current conversation to be summarized, the search will be conducted from the perspective of similar length.

[0073] Furthermore, since the purpose of summarizing is to summarize the main contents of the dialogue text, and the semantics in the dialogue text is the content to be expressed by the dialogue text, in the embodiment of the present application, when searching for the current dialogue with summary, the similarity in dialogue semantics also needs to be considered. Therefore, in the embodiment of the present application, step S102 specifically searches for dialogue texts similar to the current dialogue to be summarized in terms of dialogue length and dialogue semantics from the pre-built dialogue knowledge base as similar dialogue texts.

[0074] Optionally, specifically, N dialogue texts that are most similar to the current dialogue to be summarized in terms of dialogue length and dialogue semantics may be determined as similar dialogue texts.

[0075] Optionally, the sum of the similarities of the conversation length and the conversation semantics with the current conversation to be summarized may be used as the final similarity with the current conversation to be summarized. Alternatively, the similarity of the conversation semantics may be further considered while considering the similarity of the conversation length to select multiple similar conversation texts of the current conversation to be summarized.

[0076] Optionally, another embodiment of the present application provides a method for constructing a dialogue knowledge base, such as Figure 2 As shown, the following steps are included:

[0077] S201. Build a dialogue knowledge base using multiple sample data.

[0078] Among them, one example data includes at least one dialogue text example and a dialogue summary corresponding to the dialogue text example.

[0079] S202: Calculate the distribution statistics of the lengths of the dialogue text examples in the dialogue knowledge base.

[0080] It should be noted that, in order to facilitate bucketing and storing each dialogue text example according to length, in the embodiment of the present application, the length of each dialogue text example in the dialogue knowledge base is first distributed and counted.

[0081] S203. According to the amount of information in the dialogue text, the dialogue knowledge base is bucketed according to the text length to obtain a plurality of dialogue knowledge base length intervals.

[0082] Specifically, the text length range covered by the amount of information is divided into a conversation knowledge base length interval, that is, within a text length range, the amount of information protected is almost consistent, so the conversation knowledge base is bucketed according to the text length, thereby obtaining multiple conversation knowledge base length intervals, so that conversation text examples with similar lengths can be found more quickly in the future, that is, conversation text examples with consistent data amounts can be obtained.

[0083] Accordingly, in the embodiment of the present application, a specific implementation of step S102 is as follows: Figure 3 As shown, including:

[0084] S301. Retrieve from the dialogue knowledge base a dialogue knowledge base length interval to which the current dialogue to be summarized belongs.

[0085] Specifically, the number of characters of the current dialogue to be summarized may be determined first, and then the dialogue knowledge base length interval in which the number of characters of the current dialogue to be summarized is located may be retrieved from the dialogue knowledge base, thereby retrieving the dialogue knowledge base length interval to which it belongs.

[0086] S302. Calculate the similarity between the current dialogue to be summarized and each dialogue text example within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs, using the text vector of the current dialogue to be summarized and the text vectors of each dialogue text example within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs.

[0087] It should be noted that, in the embodiment of the present application, in order to calculate the degree of similarity between the conversation texts in terms of conversation semantics, in the embodiment of the present application, the conversation text is encoded and converted from a text format into a corresponding text vector format, so that the similarity between the conversation texts can be calculated through a similarity algorithm.

[0088] Since the text lengths of the dialogue text examples within the dialogue knowledge base length interval to which the current dialogue to be summarized belongs are similar to those of the current dialogue to be summarized, it is necessary to calculate the similarity between the current dialogue to be summarized and all dialogue text examples stored within the dialogue knowledge base length interval, thereby obtaining the similarity in dialogue semantics between the current dialogue to be summarized and the dialogue text examples within the dialogue knowledge base length interval.

[0089] S303: Obtain the top K dialogue text examples with the greatest similarity to the current dialogue to be summarized and their corresponding dialogue summaries, and obtain each similar dialogue text of the dialogue to be summarized and the dialogue summary corresponding to each similar dialogue text.

[0090] Since the greater the similarity with the current conversation to be summarized, the greater the reference value for accurately generating a conversation summary of the current text to be summarized, in an embodiment of the present application, the top K conversation text examples with the largest four degrees in Xi'an are selected as similar conversation texts of the current conversation to be summarized, and these similar conversation texts and their corresponding conversation summaries are obtained.

[0091] Optionally, considering that the conversation text examples in the current conversation knowledge base may have a low similarity with the current conversation to be summarized, when screening similar conversation texts, you can also add a condition that the similarity with the current conversation to be summarized is greater than a threshold, that is, screen the top K conversation text examples with the similarity with the current conversation to be summarized greater than the threshold and with the greatest similarity.

[0092] S103, using the current dialogue to be summarized, each similar dialogue text found and its corresponding dialogue summary, to generate model prompt information corresponding to the current dialogue to be summarized.

[0093] Optionally, by taking each similar conversation text and its corresponding conversation summary as an example and taking the current conversation to be summarized as a processing object, model prompt information can be generated so that the large model can summarize the processing object according to each example, thereby obtaining the model prompt information corresponding to the current conversation to be summarized.

[0094] Optionally, in another embodiment of the present application, a specific implementation of step S103 includes:

[0095] Each similar dialogue text and its corresponding dialogue summary are added to the model prompt information template as a reference example, and the current dialogue to be summarized is added to the model prompt information template as a target to be summarized, so as to obtain the model prompt information corresponding to the current dialogue to be summarized.

[0096] Specifically, in order to facilitate efficient and simple generation of the model prompt information of the representation, in the embodiment of the present application, the model prompt information is generated by means of a template. Therefore, each similar dialogue text and its corresponding dialogue summary can be used as a reference example, and the current dialogue to be summarized can be added to the template as the target to be summarized, so that the model prompt information corresponding to the current dialogue to be summarized can be obtained.

[0097] Optionally, the specific model prompt information template can be as follows:

[0098] “Prompt = You are an invitation expert. You need to generate a conversation summary in the following format:

[0099] [Example 1]

[0100] {Similar dialogue and summary 1}

[0101] [Example 2]

[0102] {Similar dialogue and summary 2}

[0103] [Example 3]

[0104] {Similar dialogue and summary 3}

[0105] Here’s what you need to generate a summary of your conversation:

[0106] Dialog: {Enter request dialog}

[0107] Summarize:".

[0108] Among them, the model prompt information template includes multiple columns for reference examples and columns for targets to be summarized, so that each similar dialogue text and its corresponding dialogue summary can be directly added to the corresponding column in the model prompt information template as a reference example, and the current dialogue to be summarized can be added to the corresponding column in the model prompt information template as the target to be summarized, so as to obtain the model prompt information corresponding to the current dialogue to be summarized.

[0109] For example, the current dialogue to be summarized is "Inviter: I'm from XXX. When is it convenient for you to come over to take a look today? | Driver: I don't want your car sticker, which is so big." Three similar dialogue texts and their corresponding dialogue summaries are divided into "Dialogue: Inviter: Hello, I'm from the XXX platform. | Driver: No car sticker. Summary: The driver refused to join because he didn't want the car sticker.", "Dialogue: Inviter: Master, are you ready to join XXX yet? | Driver: I don't consider the car sticker. Summary: The driver refused to join XXX because he didn't consider the car sticker.", "Dialogue: Inviter: Master, I'm from XXX. | Driver: Other companies don't want car stickers. Summary: The driver refused to join XXX because other companies don't want car stickers.", so the generated model prompt information corresponding to the current dialogue to be summarized can be as follows:

[0110] "You are an invitation expert, and you need to generate a conversation summary in the following format:

[0111] [Example 1]

[0112] Dialogue: Inviter: Hello, I am from XXX platform. | Driver: No car sticker required.

[0113] Summary: The driver refused to join XXX because he did not want the car sticker.

[0114] [Example 2]

[0115] Dialogue: Inviter: Master, are you still planning to join XXX? | Driver: I don’t consider car stickers.

[0116] Summary: The driver refused to join XXX because he did not consider the car sticker.

[0117] [Example 3]

[0118] Dialogue: Inviter: Sir, I am from XXX. | Driver: Other companies do not need car stickers.

[0119] Summary: The driver refused to join XXX because other companies did not want vehicle stickers.

[0120] Here’s what you need to generate a summary of your conversation:

[0121] Dialogue: Inviter: I am from XXX. Is it convenient for you to come over and take a look today? | Driver: I don’t want your car sticker because it is so big.

[0122] Summarize:".

[0123] S104: input the model prompt information corresponding to the current dialogue to be summarized into the big model, and generate a dialogue summary corresponding to the current dialogue to be summarized through the big model.

[0124] Specifically, the model prompt information corresponding to the current dialogue to be summarized is input into the big model. Through the model prompt information corresponding to the current dialogue to be summarized, the big model is guided to summarize the current dialogue to be summarized according to the output similar dialogue text and its corresponding summary style, thereby generating a dialogue summary corresponding to the current dialogue to be summarized.

[0125] Optionally, in the embodiment of the present application, after executing step S104, the intention of the invited person to accept the invitation is further analyzed. Figure 4 As shown, an embodiment of the present application provides a method for analyzing the intention of receiving an invitation, including:

[0126] S401: Determine the intention label of each similar dialogue text.

[0127] In the embodiment of the present application, the intention to accept the invitation is expressed by the intention tag, for example, it can include intention, no intention, fuzzy intention, etc. Since the information and semantics contained in the similar conversation are similar to the current conversation to be summarized, the intention expressed by it is usually consistent with the current conversation to be summarized. Therefore, in the embodiment of the present application, the intention tag of the current conversation to be summarized is determined by obtaining the intention tags of each similar conversation text as a reference.

[0128] Optionally, in another embodiment of the present application, a specific implementation of step S401 includes:

[0129] From the conversation knowledge base, find the intent tags of each similar conversation text.

[0130] Among them, the intention labels of similar dialogue texts are pre-matched according to the reason labels in the dialogue summary corresponding to the similar dialogue texts, and stored in the dialogue knowledge base.

[0131] Since the intention of the invitee will be expressed through his or her conversation, and is usually expressed by stating the reason, in the embodiment of the present application, the reason tags in the conversation summary corresponding to the similar conversation text are analyzed, such as "prepare materials", "no car", "consider the cost", etc., and then the corresponding intention tags are matched according to the reason tags. In addition, in order to facilitate subsequent direct use, in the embodiment of the present application, after the intention tags of the conversation text are analyzed and obtained, they will be directly stored in the conversation knowledge base.

[0132] S402. According to the majority voting principle, the intention tag with the largest number among the intention tags of the similar dialogue texts is used as the intention tag of the dialogue summary corresponding to the current dialogue to be summarized.

[0133] Since the larger the number of a certain intention tag, the greater the probability that the conversation summary corresponding to the current conversation to be summarized belongs to the intention tag, in an embodiment of the present application, the intention tag with the largest number is used as the intention tag of the conversation summary corresponding to the current conversation to be summarized.

[0134] Accordingly, in the embodiment of the present application, after executing step S402, the following steps are further included:

[0135] The conversation summary and its intention label corresponding to the current conversation to be summarized are output as the intelligent summary of the current conversation to be summarized.

[0136] The embodiment of the present application provides a method for generating a dialogue summary, which obtains the current dialogue to be summarized, and based on the dialogue length and dialogue semantics, finds out multiple similar dialogue texts of the current dialogue to be summarized and the dialogue summaries corresponding to each of the similar dialogue texts from a pre-constructed dialogue knowledge base, so that by finding dialogue texts similar to the current dialogue to be summarized in length and semantics, the information contained and the dialogue texts with similar semantics to the current dialogue to be summarized can be obtained, so that the dialogue summary of the current dialogue to be summarized can be made with reference to the similar dialogue texts and their corresponding dialogue summaries. Then, the model prompt information corresponding to the current dialogue to be summarized is generated by using the current dialogue to be summarized, each of the similar dialogue texts found out and their corresponding dialogue summaries, so that dynamic model prompt information can be constructed to more accurately summarize the current dialogue to be summarized. Finally, the model prompt information corresponding to the current dialogue to be summarized is input into the large model, and the dialogue summary corresponding to the current dialogue to be summarized is generated by the large model, that is, the large model summarizes the current dialogue to be summarized with reference to the similar dialogue texts and their corresponding dialogue summaries under the guidance of the model prompt information, thereby effectively ensuring the accuracy of the dialogue summary.

[0137] Another embodiment of the present application provides a conversation summary generating device, such as Figure 5 As shown, including:

[0138] The conversation acquisition unit 501 is used to acquire the current conversation to be summarized.

[0139] The conversation search unit 502 is used to search for multiple similar conversation texts of the current conversation to be summarized and the conversation summaries corresponding to the respective similar conversation texts from a pre-built conversation knowledge base based on the conversation length and conversation semantics.

[0140] The generating unit 503 is used to generate model prompt information corresponding to the current dialogue to be summarized by using the current dialogue to be summarized, each similar dialogue text found and its corresponding dialogue summary.

[0141] The summarizing unit 504 is used to input the model prompt information corresponding to the current dialogue to be summarized into the large model, and generate a dialogue summary corresponding to the current dialogue to be summarized through the large model.

[0142] Optionally, in another embodiment of the present application, the conversation summary generating device further includes:

[0143] The construction unit is used to construct a dialogue knowledge base using multiple sample data, wherein one sample data at least includes a dialogue text sample and a dialogue summary corresponding to the dialogue text sample.

[0144] The statistical unit is used to perform distribution statistics on the length of each dialogue text example in the dialogue knowledge base.

[0145] The partitioning unit is used to bucket the dialogue knowledge base according to the length of the text according to the amount of information in the dialogue text, so as to obtain multiple dialogue knowledge base length intervals.

[0146] Optionally, in the conversation summary generating device provided in another embodiment of the present application, the conversation searching unit includes:

[0147] The retrieval unit is used to retrieve the dialogue knowledge base length interval to which the current dialogue to be summarized belongs from the dialogue knowledge base.

[0148] The calculation unit is used to calculate the similarity between the current dialogue to be summarized and each dialogue text example within the length interval of the dialogue knowledge base to which it belongs, using the text vector of the current dialogue to be summarized and the text vector of each dialogue text example within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs.

[0149] The extraction unit is used to obtain the top K dialogue text examples with the greatest similarity to the current dialogue to be summarized and their corresponding dialogue summaries, and obtain each similar dialogue text of the dialogue to be summarized and the dialogue summary corresponding to each similar dialogue text.

[0150] Optionally, in a conversation summary generating device provided in another embodiment of the present application, the generating unit includes:

[0151] A generation subunit is used to add each similar dialogue text and its corresponding dialogue summary as a reference example to the model prompt information template, and add the current dialogue to be summarized as a target to be summarized to the model prompt information template to obtain the model prompt information corresponding to the current dialogue to be summarized.

[0152] Optionally, in another embodiment of the present application, the conversation summary generating device further includes:

[0153] The label determination unit is used to determine the intention label of each similar dialogue text.

[0154] The voting unit is used to, according to the majority voting principle, use the intention label with the largest number among the intention labels of each similar dialogue text as the intention label of the dialogue summary corresponding to the current dialogue to be summarized.

[0155] Optionally, in the conversation summary generating device provided in another embodiment of the present application, the label determining unit includes:

[0156] The label search unit is used to search for the intention labels of each similar dialogue text from the dialogue knowledge base. The intention labels of the similar dialogue texts are pre-matched according to the reason labels in the dialogue summary corresponding to the similar dialogue texts and stored in the dialogue knowledge base.

[0157] Optionally, in another embodiment of the present application, the conversation summary generating device further includes:

[0158] The output unit is used to output the conversation summary and its intention label corresponding to the current conversation to be summarized as the intelligent summary of the current conversation to be summarized.

[0159] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, which will not be repeated here.

[0160] Another embodiment of the present application provides an electronic device, such as Figure 6 As shown, including:

[0161] Memory 601 and processor 602 .

[0162] Among them, the memory is used to store programs.

[0163] The processor 602 is used to execute the program stored in the memory 601. When the program is executed by the processor 602, it is specifically used to implement the conversation summary generation method provided in any one of the above embodiments.

[0164] Another embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed by an electronic device, it is used to implement the conversation summary generation method provided in any of the above embodiments.

[0165] Computer storage media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0166] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0167] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating a dialogue summary, characterized in that: include: Get the current conversation to be summarized; Based on the conversation length and conversation semantics, searching from a pre-built conversation knowledge base, a plurality of similar conversation texts of the current conversation to be summarized and conversation summaries corresponding to each of the similar conversation texts; Generate model prompt information corresponding to the current conversation to be summarized by using the current conversation to be summarized, each of the similar conversation texts found and their corresponding conversation summaries; The model prompt information corresponding to the current dialogue to be summarized is input into the large model, and a dialogue summary corresponding to the current dialogue to be summarized is generated by the large model.

2. The method according to claim 1, characterized in that Also includes: The conversation knowledge base is constructed using a plurality of example data; wherein one of the example data at least includes a conversation text example and a conversation summary corresponding to the conversation text example; Performing distribution statistics on the lengths of each of the dialogue text examples in the dialogue knowledge base; According to the amount of information in the dialogue text, the dialogue knowledge base is bucketed according to the text length to obtain multiple dialogue knowledge base length intervals.

3. The method according to claim 2, characterized in that The method of searching, based on the conversation length and the conversation semantics, from a pre-built conversation knowledge base, a plurality of similar conversation texts of the current conversation to be summarized and conversation summaries corresponding to each of the similar conversation texts, includes: Retrieving from the conversation knowledge base a conversation knowledge base length interval to which the current conversation to be summarized belongs; Calculate the similarity between the current dialogue to be summarized and each of the dialogue text examples within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs, using the text vector of the current dialogue to be summarized and the text vectors of each of the dialogue text examples within the length interval of the dialogue knowledge base to which the current dialogue to be summarized belongs; The first K dialogue text examples with the greatest similarity to the current dialogue to be summarized and their corresponding dialogue summaries are obtained to obtain each of the similar dialogue texts of the dialogue to be summarized and the dialogue summaries corresponding to each of the similar dialogue texts.

4. The method according to claim 1, characterized in that The step of using the current conversation to be summarized, each of the similar conversation texts found and their corresponding conversation summaries to generate model prompt information corresponding to the current conversation to be summarized includes: Each of the similar dialogue texts and its corresponding dialogue summary is added to the model prompt information template as a reference example, and the current dialogue to be summarized is added to the model prompt information template as a target to be summarized, so as to obtain the model prompt information corresponding to the current dialogue to be summarized.

5. The method according to claim 1, characterized in that After inputting the model prompt information corresponding to the current dialogue to be summarized into the large model and generating the dialogue summary corresponding to the current dialogue to be summarized by the large model, the method further includes: Determining the intent tags of each of the similar dialogue texts; According to the majority voting principle, the intention tag with the largest number among the intention tags of the similar conversation texts is used as the intention tag of the conversation summary corresponding to the current conversation to be summarized.

6. The method according to claim 5, characterized in that The determining of the intention labels of the respective similar dialogue texts includes: From the conversation knowledge base, find out the intention tags of each of the similar conversation texts; wherein the intention tags of the similar conversation texts are pre-matched based on the reason tags in the conversation summaries corresponding to the similar conversation texts, and stored in the conversation knowledge base.

7. The method according to claim 5, characterized in that After the intention tag with the largest number among the intention tags of the similar dialogue texts is used as the intention tag of the dialogue summary corresponding to the current dialogue to be summarized according to the majority voting principle, the method further includes: The conversation summary and its intention label corresponding to the current conversation to be summarized are output as the intelligent summary of the current conversation to be summarized.

8. A conversation summary generating device, characterized in that: include: A dialogue acquisition unit, used to acquire the current dialogue to be summarized; A conversation search unit, configured to search, based on the conversation length and conversation semantics, from a pre-built conversation knowledge base for a plurality of similar conversation texts of the current conversation to be summarized and conversation summaries corresponding to the respective similar conversation texts; A generating unit, configured to generate model prompt information corresponding to the current conversation to be summarized by using the current conversation to be summarized, each of the similar conversation texts found and their corresponding conversation summaries; The summarizing unit is used to input the model prompt information corresponding to the current dialogue to be summarized into the large model, and generate a dialogue summary corresponding to the current dialogue to be summarized through the large model.

9. An electronic device, characterized in that: include: Memory and processor; Wherein, the memory is used to store programs; The processor is used to execute the program, and when the program is executed, it is specifically used to implement the conversation summary generation method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, is used to implement the conversation summary generation method as described in any one of claims 1 to 7.