Information processing method and device, equipment, storage medium and program product

Abstract

The invention discloses an information processing method and device, equipment, a storage medium and a program product, and relates to the technical field of natural language processing, and the method comprises the steps: carrying out the voice recognition of a call voice, obtaining a call transcription text, determining a call type representing a call purpose based on the call transcription text, and storing the call type. And performing summary extraction on the call transcription text based on the call type to obtain a call summary in a format corresponding to the call type. Based on the method and the device, the conversation summary in different formats can be obtained according to different conversation types, and the diversity of the formats of the generated conversation summary is improved, so that the auxiliary effect of the conversation summary on a user is improved.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular, to an information processing method, apparatus, device, storage medium, and program product. Background Art

[0002] With the rapid development of communication technology, telephone conferences, video conferences, etc. have become important ways for people to communicate in work and life. In some scenarios, it is necessary to organize the call content after the meeting to obtain information such as call minutes.

[0003] The call minutes generated by the current minutes generation method have a single format and poor auxiliary effect on users. Summary of the Invention

[0004] In view of the above problems, this application provides an information processing method, apparatus, device, storage medium, and program product to improve the auxiliary effect of call minutes on users. The specific solutions are as follows:

[0005] The first aspect of this application provides an information processing method, including:

[0006] Performing speech recognition on the call voice to obtain a call transcription text;

[0007] Determining the call type based on the call transcription text; different call types correspond to different call purposes;

[0008] Performing minutes extraction on the call transcription text based on the call type to obtain call minutes in a format corresponding to the call type.

[0009] In a possible implementation, the determining the call type based on the call transcription text includes:

[0010] Dividing the call transcription text into at least one text segment; each text segment contains consecutive content in the call transcription text; the length of each text segment is less than or equal to a preset length;

[0011] Determining the call type based on at least one text segment;

[0012] Among them, determining the call type based on one text segment includes: adding the one text segment to the first call slot of the first prompt word template to obtain a first prompt word; the first prompt word template also includes a first task instruction, and the first task instruction is: judging the call type based on the text segment in the first call slot; inputting the first prompt word into a large model to obtain the call type output by the large model based on the text segment in the first call slot;

[0013] Determining the call type based on at least two text segments includes: for each text segment, adding the text segment to the first call slot of the first prompt template to obtain the first prompt word corresponding to the text segment; inputting the first prompt word corresponding to the text segment into the large model to obtain the call type corresponding to the text segment output by the large model based on the text segment in the first call slot; determining the final call type based on the call types corresponding to each text segment in the at least two text segments.

[0014] In a possible implementation, determining the final call type based on the call types corresponding to each text segment in the at least two text segments includes:

[0015] If the call types corresponding to each text segment in the at least two text segments are the same, determining the call type corresponding to any one text segment as the final call type;

[0016] If there are at least two text segments in the at least two text segments whose corresponding call types are different, determining the general type as the final call type.

[0017] In a possible implementation, the speaker information and its corresponding speech content are recorded in the call transcription text in the chronological order of the speaker's speech; each speaker information and its corresponding speech content are marked as a paragraph; when dividing the call transcription text into at least two text segments:

[0018] Each text segment includes one or consecutive multiple paragraphs in the call transcription text;

[0019] Two adjacent text segments do not cover the same paragraph, and two adjacent text segments cover consecutive multiple paragraphs in the call transcription text.

[0020] In a possible implementation, extracting a summary of the call transcription text based on the call type includes:

[0021] For each text segment, adding the text segment to the second call slot of the second prompt template corresponding to the call type to obtain the second prompt word corresponding to the text segment; the second prompt template also includes a second task instruction, the call type, and its corresponding summary template, and the second task instruction is: generate a summary of the text segment in the second call slot according to the summary template; different summary templates correspond to different summary formats;

[0022] Inputting the second prompt word corresponding to the text segment into the large model to obtain the summary of the text segment generated by the large model according to the summary template;

[0023] In the case of dividing the call transcription text into at least two text segments, the summaries of each text segment are integrated to obtain a complete call summary in a format corresponding to the call type.

[0024] In one possible implementation, for each text segment, adding the text segment to the second call slot of the second prompt template corresponding to the call type includes:

[0025] Determine the length range to which the length of the text segment belongs;

[0026] Add the text segment to the second call slot of the second prompt template corresponding to the call type and the length range; for different length ranges, the summary templates corresponding to the call type in the second prompt template are different.

[0027] In one possible implementation, the integrating the summaries of each text segment to obtain a complete call summary in a format corresponding to the call type includes:

[0028] Concatenate the summaries of each text segment into one or more concatenated summaries; each concatenated summary is obtained by concatenating the summaries of at least two consecutive text segments; different concatenated summaries cover the summaries of different text segments;

[0029] For each concatenated summary, add the concatenated summary to the summary segment slot of the third prompt template corresponding to the call type to obtain the third prompt corresponding to the concatenated summary; the third prompt template also includes the call type and a third task instruction, and the third task instruction is: sort out the concatenated summaries in the summary segment slot to obtain a complete call summary;

[0030] Input the third prompt corresponding to the concatenated summary into the large model to obtain the integration result of the concatenated summary generated by the large model;

[0031] In the case of concatenating the summaries of each text segment into one concatenated summary, the integration result of the concatenated summary is a complete call summary in a format corresponding to the call type;

[0032] In the case of concatenating the summaries of each text segment into multiple concatenated summaries, concatenate the integration results of each concatenated summary and add them to the third prompt template to obtain the third prompt corresponding to the concatenated integration result; input the third prompt corresponding to the concatenated integration result into the large model to obtain a complete call summary in a format corresponding to the call type generated by the large model.

[0033] In one possible implementation, it further includes:

[0034] Perform to-do extraction on the call transcription text to obtain the to-do items of each speaker and the retrieval sentences corresponding to each to-do item;

[0035] The retrieval sentence corresponding to each to-do item is the sentence in the call transcription text corresponding to the to-do item.

[0036] In a possible implementation, performing to-do extraction on the call transcription text includes:

[0037] Divide the call transcription text into at least one text segment; each text segment contains consecutive content in the call transcription text; the length of each text segment is less than or equal to a preset length;

[0038] For each text segment, add the text segment to the third call slot of the fourth prompt template to obtain the fourth prompt corresponding to the text segment; the fourth prompt template also includes a fourth task instruction, and the fourth task instruction is: Extract the to-do items that each speaker needs to execute and the retrieval sentences corresponding to each to-do item from the text segment in the third call slot according to the specified requirements;

[0039] Input the fourth prompt corresponding to the text segment into the large model to obtain the to-do items of each speaker corresponding to the text segment generated by the large model and the retrieval sentences corresponding to each to-do item;

[0040] In the case of dividing the call transcription text into at least two text segments, merge the to-do items and corresponding retrieval sentences of the same speaker corresponding to each text segment to obtain the to-do items of each speaker corresponding to the call transcription text and the retrieval sentences corresponding to each to-do item.

[0041] The second aspect of this application provides an information processing device, including:

[0042] An identification module for performing speech recognition on call speech to obtain a call transcription text;

[0043] A type determination module for determining the call type based on the call transcription text; different call types correspond to different call purposes;

[0044] A minutes extraction module for performing minutes extraction on the call transcription text based on the call type to obtain a call minutes in a format corresponding to the call type.

[0045] The third aspect of this application provides a computer program product, including computer-readable instructions, which when running on an electronic device, enable the electronic device to implement the information processing method in the first aspect or any implementation manner of the first aspect.

[0046] A fourth aspect of the present application provides an electronic device, including at least one processor and a memory connected to the processor, where:

[0047] The memory is used to store a computer program;

[0048] The processor is used to execute the computer program so that the electronic device can implement the information processing method of the above first aspect or any implementation manner of the first aspect.

[0049] A fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the information processing method of the above first aspect or any implementation manner of the first aspect.

[0050] By means of the above technical solutions, the information processing method, device, equipment, storage medium and program product provided by the present application perform speech recognition on the call voice to obtain a call transcription text, determine the call type representing the call purpose based on the call transcription text, and perform minutes extraction on the call transcription text based on the call type to obtain a call minutes in a format corresponding to the call type. Based on the present application, different formats of call minutes can be obtained according to different call types, improving the diversity of the generated call minutes formats, and thus improving the auxiliary effect of the call minutes on users. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Combined with the drawings and referring to the following specific implementation manners, the above and other features, advantages and aspects of each embodiment of the present disclosure will become more obvious. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic, and the original elements and elements are not necessarily drawn to scale.

[0052] Figure 1 is a flowchart of an implementation of the information processing method provided by the present application;

[0053] Figure 2 is a flowchart of an implementation of determining the call type based on the call transcription text provided by the present application;

[0054] Figure 3 is an example diagram of the call transcription text with paragraph marks added provided by the present application;

[0055] Figure 4 is a flowchart of an implementation of performing minutes extraction on the call transcription text based on the call type provided by the present application;

[0056] Figure 5 is another flowchart of an implementation of performing minutes extraction on the call transcription text based on the call type provided by the present application;

[0057] Figure 6 A flowchart of an implementation for integrating the minutes of each text segment provided in this application;

[0058] Figure 7 An example of integrating the minutes of two text segments provided in this application to obtain a complete call record in a format corresponding to the call type;

[0059] Figure 8a An example diagram of the to - do items of both parties in a call provided in this application;

[0060] Figure 8b An example of the to - do items and retrieval sentences extracted provided in this application;

[0061] Figure 9 A flowchart of an implementation for extracting to - do items from call record text provided in this application;

[0062] Figure 10 An example of merging the to - do items of two text segments provided in this application to obtain the final to - do items;

[0063] Figure 11 A schematic structural diagram of an information processing device provided in this application;

[0064] Figure 12 A schematic structural diagram of an electronic device provided in this application. Detailed implementation manners

[0065] The embodiments of this application will be described below with reference to the accompanying drawings in the embodiments of this application. The terms used in the embodiment part of this application are only used to explain the specific embodiments of this application, rather than to limit this application.

[0066] The embodiments of this application will be described below with reference to the accompanying drawings. Those of ordinary skill in the art will know that with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0067] The terms "first", "second", etc. in the specification, claims and the above - mentioned accompanying drawings of this application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances, which is only a way of distinguishing objects with the same attributes when describing the embodiments of this application. In addition, the terms "include" and "have" and any of their variations are intended to cover non - exclusive inclusion, so that a process, method, system, product or device including a series of units does not have to be limited to those units, but may include other units not clearly listed or inherent to these process, methods, products or devices.

[0068] In order to improve the intelligence of call summary generation, increase the diversity of the generated call summaries, and improve the auxiliary effect of the generated call summaries on users, this application is proposed.

[0069] The information processing method provided by the embodiments of this application can be used in an electronic device (for the convenience of description and distinction, denoted as the first electronic device), which can establish a call connection with at least one other electronic device (for the convenience of description and distinction, denoted as the second electronic device), so that the first user of the first electronic device can have a voice call with the second user of the second electronic device.

[0070] The call connection between the first electronic device and the second electronic device can be established through a mobile communication network (such as, 4G network, 5G network, 6G network, etc.), or can also be established through the Internet.

[0071] As Figure 1 shown, it is a flowchart of an implementation of the information processing method provided by the embodiments of this application, which may include:

[0072] Step S101: Perform speech recognition on the call voice to obtain a call transcription text.

[0073] Optionally, voice recording can be performed during the voice call between the first electronic device and the second electronic device, and after the call ends, perform speech recognition on the recorded call voice to obtain a call transcription text; or, during the voice call between the first electronic device and the second electronic device, perform speech recognition in real time, and after the call ends, obtain a call transcription text.

[0074] Step S102: Determine the call type based on the call transcription text; different call types correspond to different call purposes.

[0075] In this application, as an example, the call type may include but is not limited to the following: notification type, consultation type, invitation type, discussion type, reservation type, marketing type, and general type, etc.

[0076] Among them, a notification-type call is a call for one party of the call to notify the other party of some events or requirements.

[0077] A consultation-type call is a call for one party of the call to raise and consult a series of questions or events to the other party.

[0078] An invitation-type call is a call for one party of the call to invite the other party to participate in an activity or do something together.

[0079] A discussion-type call is a call for both parties of the call to exchange opinions, share views and information around a specific topic.

[0080] A reservation call is a call in which one party to the call reserves a certain service or resource, etc. for the other party.

[0081] A marketing call is a call in which one party to the call promotes a certain service or resource, etc. to the other party.

[0082] A general call is a call type other than the above several call types.

[0083] The call type of the call transcription text can be identified by the large model for at least part of the content to determine the call type corresponding to the call transcription text.

[0084] In the embodiments of the present application, the large model can be a large language model (LLM) or a multimodal large model (MLLMs).

[0085] The large model can be a generative model. As an example, it can include, but is not limited to, models with the following architectures: Transformer, PaLM (Pathways Language Model), T5 (Text-to-Text Transfer Transformer), etc.

[0086] Step S103: Extract a summary of the call transcription text based on the call type to obtain a call summary in a format corresponding to the call type.

[0087] When extracting a summary of the call transcription text in the present application, the call type is considered. According to the different call types, the formats of the obtained call summaries are different, increasing the format diversity of the generated call summaries.

[0088] As an example, for a call summary of the notice type, the three modules of the notice content, the response of the other party, and the final conclusion should be written. The call summary of the consultation type, on the other hand, revolves around each question and writes out the consultation questions and answers, etc.

[0089] The information processing method provided by the present application performs speech recognition on the call voice to obtain a call transcription text, determines the call type representing the call purpose based on the call transcription text, and extracts a summary of the call transcription text based on the call type to obtain a call summary in a format corresponding to the call type. Based on the present application, different formats of call summaries can be obtained according to different call types, improving the intelligence of call summary generation, increasing the format diversity of the generated call summaries, and thus improving the auxiliary effect of the call summary on users.

[0090] In an optional embodiment, a flowchart of one implementation of determining the call type based on the call transcription text is asFigure 2 As shown, it may include:

[0091] Step S201: Divide the call transcription text into at least one text segment; each text segment contains consecutive content in the call transcription text; the length of each text segment is less than or equal to a preset length.

[0092] Optionally, it can be determined whether to divide the call transcription text into one text segment or multiple text segments according to the length of the call transcription text.

[0093] If the length of the call transcription text is less than or equal to the preset length, divide the call transcription text into one text segment, that is, the call transcription text itself is used as one text segment.

[0094] If the length of the call transcription text is greater than the preset length, then divide the call transcription text into multiple text segments. Each text segment contains consecutive content in the call transcription text, and the adjacent two text segments constitute consecutive content in the call transcription text. The length of each text segment is less than or equal to the preset length.

[0095] The length of the call transcription text can be the number of words in the call transcription text, and the preset length can be the preset number of words.

[0096] Step S202: Determine the call type based on at least one text segment.

[0097] Optionally, in the case of dividing the call transcription text into one text segment, determine the call type based on this one text segment.

[0098] In the case of dividing the call transcription text into multiple (i.e., at least two) text segments, the call type can be determined based on only one of the text segments. This one text segment can be the first text segment (the first text segment is the text segment containing the first sentence in the call transcription text), or the last text segment, or it can be a certain intermediate text segment.

[0099] In the case of dividing the call transcription text into multiple (i.e., at least two) text segments, the call type can also be determined based on at least two (for the sake of narration and distinction, denoted as M, where M is an integer greater than 1) text segments. These M text segments can be some of the text segments obtained by division, or all of the text segments obtained by division.

[0100] As an example, regardless of whether the call transcription text is divided into one text segment or multiple text segments, one implementation method for determining the call type based on one text segment can be:

[0101] Add this text fragment to the first call slot of the first prompt template to obtain the first prompt.

[0102] The first prompt template also includes a first task instruction, which is to determine the call type based on the text fragment in the first call slot. As an example, the first prompt template could be:

[0103] "Please determine the type of this call, whether it is a consultation type, a notification type, or other types; Notification type: I (the other party) actively notify the other party (me) of some events or requirements; Consultation type: I (the other party) ask a series of questions or events to the other party (me). The output format is: "Call type: xx".

[0104] The output reference examples are as follows:

[0105] Call type: Consultation type.

[0106] Call type: Notification type.

[0107] Call type: General type.

[0108] Input call: "

[0109] In the above example of the first prompt template, after "Input call:", there is the first call slot for filling the text fragment.

[0110] Input the first prompt into the large model to obtain the call type output by the large model based on the text fragment in the first call slot.

[0111] Optionally, in the case of dividing the call transcription text into multiple text fragments, one implementation of determining the call type based on at least two text fragments can be: determine the corresponding call type for each text fragment among the at least two text fragments, and determine the final call type based on the call types corresponding to each text fragment among the at least two text fragments.

[0112] Optionally, if the call types corresponding to each text fragment among the at least two text fragments are the same, then the call type corresponding to any one text fragment is the finally determined call type.

[0113] If there are at least two text fragments among the at least two text fragments whose corresponding call types are different, determine the general type as the final call type. That is to say, if there are at least two text fragments whose corresponding call types are different, unify the call types corresponding to each text fragment into the general type, that is, determine the general type as the final call type.

[0114] As an example, for each of the at least two text segments, one implementation of determining the call type corresponding to the text segment based on the text segment may be:

[0115] Add the text segment to the first call slot of the above first prompt word template to obtain the first prompt word corresponding to the text segment.

[0116] Input the first prompt word corresponding to the text segment into the large model to obtain the call type corresponding to the text segment output by the large model based on the text segment in the first call slot.

[0117] In an optional embodiment, the speaker information and its corresponding speech content are recorded in the call transcription text in the chronological order of the speaker's speech; each speaker information and its corresponding speech content are marked as a paragraph.

[0118] As Figure 3 shown, it is an example diagram of the call transcription text with paragraph markers added provided by the embodiments of the present application. In this example, the paragraph marker is "(segid: serial number)"; "<Me>" and "<Other Party>" are speaker information, "<Me>" represents the user of the first electronic device, and "<Other Party>" represents the user of the second electronic device. The speaker information can be distinguished from different users through voiceprint recognition during the process of speech recognition, and then the speaker information is determined. After obtaining the call transcription text, a paragraph marker can be added before each speaker information in sequence from front to back to mark different paragraphs.

[0119] Based on this, when dividing the call transcription text into at least two text segments:

[0120] Each text segment includes one paragraph or consecutive multiple paragraphs in the call transcription text.

[0121] Two adjacent text segments do not cover the same paragraph, and two adjacent text segments cover consecutive multiple paragraphs in the call transcription text.

[0122] That is to say, each paragraph in each text segment is a complete paragraph, and no two text segments include the same paragraph.

[0123] Optionally, the call transcription text can be divided into semantic groups by the large model, so as to divide the call transcription text into at least two text segments. As an example, the call transcription text can be added to the fourth call slot of the fifth prompt word template to obtain the fifth prompt word; the fifth prompt word template also includes a fifth task instruction, and the fifth task instruction is: divide the call transcription text in the fourth call slot into different chapters according to the specified requirements. As an example, the fifth prompt word template can be:

[0124] "Please divide the following call transcription text into different chapters and list the starting and ending paragraph IDs included in each chapter. Note that the number of words in each chapter does not exceed the preset number of words. The ID of the current paragraph is given by the '(segid:)' tag before each paragraph in the call transcription text. It is required that there is no overlap of the same paragraphs between chapters, and the starting and ending paragraph IDs of adjacent chapters cover all the paragraphs in the entire call.

[0125] Output format:

[0126] {[[segidA,segidB],[segidC,segidD],[segidE,segidF]]}

[0127] The input call is as follows:

[0128] In the example of the above fifth prompt word template, after "The input call is as follows:", there is the fourth call slot for filling the call transcription text. The "chapter" mentioned in the above example is the text fragment.

[0129] Input the fifth prompt word into the large model to obtain the chapter list output by the large model.

[0130] As an example, the chapter list output by the large model can be data in the following format:

[0131] "[[0,19],[20,36],[37,72],[73,108],[109,152],[153,213]]"

[0132] Among them, [0,19] represents the first chapter (i.e., the first text fragment), which consists of 20 paragraphs with paragraph IDs from 0 to 19 in the call transcription text, [20,36] represents the second chapter (i.e., the second text fragment), which consists of 17 paragraphs with paragraph IDs from 20 to 36 in the call transcription text, and so on.

[0133] Optionally, the call transcription text can be divided into at least two text segments according to preset rules. As an example, first calculate how many text segments the call transcription text needs to be divided into according to the preset length of each text segment. Assume that it is determined that the call transcription text needs to be divided into N text segments (obtained by dividing the length of the call transcription text by the preset length and rounding up the quotient), then calculate the average length of each of the N text segments (obtained by dividing the length of the call transcription text by N), and then traverse each paragraph in the call transcription text in chronological order. For each paragraph traversed, calculate the total length of this paragraph and the paragraphs that have not been divided into text segments before it. If the total length is equal to the preset length, divide this paragraph and the paragraphs that have not been divided into text segments before it into a text segment, and then traverse the next paragraph; if the total length is greater than the preset length, divide the paragraphs that have not been divided into text segments before this paragraph into a text segment, and then traverse the next paragraph; if the total length is less than the preset length, continue to traverse the next paragraph, and so on until all paragraphs in the call transcription text are divided into text segments. Based on this, the length of each text segment can be made relatively uniform, and there is no splitting of the text within the paragraphs between the text segments.

[0134] In an optional embodiment, in the case where the call transcription text is divided into one text segment, a flowchart of an implementation of extracting a summary of the call transcription text based on the call type is as Figure 4 shown, and may include:

[0135] Step S401: Add the text segment to the second call slot of the second prompt word template corresponding to the call type to obtain the second prompt word corresponding to the text segment.

[0136] The second prompt word template also includes a second task instruction, the call type, and its corresponding summary template. The second task instruction is: Generate a summary of the text segment in the second call slot according to the summary template.

[0137] Different summary templates correspond to different summary formats.

[0138] As an example, the summary template corresponding to the general category may be:

[0139] "In this call, I (discussed, informed, communicated, consulted, requested, reminded...) the other party...

[0140] 1) Subtitle: Content

[0141] 2) Subtitle: Content"

[0142] As an example, the summary template corresponding to the discussion category may be:

[0143] "In this call, I discussed with the other party about...

[0144] (1) (Subtitle):...

[0145] Conclusion:

[0146] (2) (Subtitle):...

[0147] Conclusion:

[0148] (3) (Subtitle):...

[0149] Conclusion: "

[0150] As an example, the corresponding minutes template for consultation can be:

[0151] "In this call, I consulted the other party about..., and the events mainly include the following:

[0152] 1) Consultation about...

[0153] The other party replied:...

[0154] 2) Consultation about...

[0155] The other party replied:...

[0156] At the same time, the other party consulted me about... events, which mainly include the following:

[0157] 1) Consultation about...

[0158] I replied:...

[0159] 2) Consultation about...

[0160] I replied:..."

[0161] As an example, the corresponding minutes template for invitation can be:

[0162] "In this call, I invited the other party to participate in..., with the theme of...

[0163] 1) Event:...

[0164] 2) Participants:...

[0165] 3) Time:...

[0166] 4) Transportation method:...

[0167] 5) Meeting place:..."

[0168] As an example, the corresponding minutes template for reservation can be:

[0169] "In this call, I reserved... from the other party

[0170] 1) Time: ……

[0171] 2) Place: ……

[0172] 3) Precautions: ……”

[0173] As an example, the corresponding minutes template for marketing can be:

[0174] "In this phone call, I recommended the XXXX event / solution / product to the other party, and the content is:

[0175] 1) Name of the event / solution / product: ……

[0176] 2) Introduction of the event / solution / product: amount, key points, features

[0177] 3) Comparison with competitors of the event / solution / product: Superior to XXXXX;

[0178] 4) Risks of the event / solution / product: ……

[0179] The other party indicated... (intention), and agreed on:

[0180] 1) Time: "

[0181] Step S402: Input the second prompt word corresponding to the text segment into the large model to obtain the minutes of the text segment generated by the large model according to the minutes template, and the minutes are the complete call minutes extracted from the call transcription text.

[0182] In an optional embodiment, in the case of dividing the call transcription text into at least two text segments, another implementation flowchart of extracting minutes from the call transcription text based on the call type is as Figure 5 shown, and may include:

[0183] Step S501: For each text segment, add the text segment to the second call slot of the second prompt word template corresponding to the call type to obtain the second prompt word corresponding to the text segment.

[0184] The second prompt word template also includes a second task instruction, the call type, and its corresponding minutes template. The second task instruction is: Generate the minutes of the text segment in the second call slot according to the minutes template.

[0185] Different minutes templates correspond to different minutes formats.

[0186] The minutes templates corresponding to different call types can be referred to in the foregoing embodiments and will not be elaborated here.

[0187] Step S502: Input the second prompt word corresponding to the text segment into the large model to obtain the summary of the text segment generated by the large model according to the minutes template.

[0188] Step S503: Integrate the information of the summaries of each text segment to obtain a complete call minutes in the format corresponding to the call type, and this minutes is the complete call minutes extracted from the call transcription text.

[0189] The summaries of adjacent text segments are not coherent, and there may be the same or similar content. Integrating the information of the summaries of each text segment means fusing each text segment into a coherent and non-redundant complete call minutes.

[0190] Optionally, the large model can be used to integrate the information of the summaries of each text segment to obtain a complete call minutes in the format corresponding to the call type.

[0191] In an actual call scenario, the length of the call transcription text is very random, with long and short lengths. The information points contained in calls of different lengths may be different. For example, for a shorter call, the minutes can be generated by summarizing it in one sentence, and for a longer call, the minutes can be generated in a summary sentence and a point-by-point format, so that the generated minutes are more concise and reasonable. Based on this, in the Figure 4 and Figure 5 illustrated embodiments, each call type can correspond to multiple second prompt word templates, and different second prompt word templates under the same call type correspond to different length ranges of text segments. For different length ranges, the minutes templates corresponding to the call type in the second prompt word templates are different (for example, the defined length of the minutes in the minutes template is different, the number of listed points is different, etc.).

[0192] Correspondingly, for each text segment, one implementation manner of filling the text segment into the second call slot of the second prompt word template corresponding to the call type can be:

[0193] Determine the length range to which the length of the text segment belongs.

[0194] Fill the text segment into the second call slot of the second prompt word template corresponding to the call type and the determined length range.

[0195] As an example, in the scenario where the call type is a general type, according to the different lengths of the call transcription text, the second prompt word template can be:

[0196] If the number of words in the text segment is less than 81, the corresponding second prompt word template can be:

[0197] "The following is a general type of call record with a word count of 10 - 80. Please generate a call summary according to the following template: In this call, I (the other party) (discussed, informed, communicated, consulted, requested, reminded....) the other party (me) (discussed, informed, communicated, consulted, requested, reminded...). Please note that only one sentence needs to be generated, and the total word count of the summary should not exceed 60% of the input call word count. Input call:"

[0198] If the word count of the text fragment is greater than 80 words and less than 251 words, the corresponding second prompt word template can be:

[0199] "The following is a general type of call record with a word count of 80 - 250. Please generate a call summary according to the following template: In this call, I (the other party) (discussed, informed, communicated, consulted, requested, reminded....) the other party (me) (discussed, informed, communicated, consulted, requested, reminded...). 1) xx 2) xx; Please note that the number of bullet points should not exceed 3, each bullet point should not exceed 50 words, and the total word count of the summary should not exceed 150 words. Input call:"

[0200] If the word count of the text fragment is greater than 250 words and less than 451 words, the corresponding second prompt word template can be:

[0201] "The following is a general type of call record with a word count of 250 - 450. Please generate a call summary according to the following template: In this call, I (the other party) (discussed, informed, communicated, consulted, requested, reminded....) the other party (me) (discussed, informed, communicated, consulted, requested, reminded...). 1) xx 2) xx; Please note that the number of bullet points should not exceed 4, each bullet point should not exceed 50 words, and the total word count of the summary should not exceed 220 words. Input call:"

[0202] If the word count of the text fragment is greater than 450 words and less than 801 words, the corresponding second prompt word template can be:

[0203] "The following is a general type of call record with a word count of 450 - 800. Please generate a call summary according to the following template: In this call, I (the other party) (discussed, informed, communicated, consulted, requested, reminded....) the other party (me) (discussed, informed, communicated, consulted, requested, reminded...). 1) xx 2) xx; Please note that the number of bullet points should not exceed 5, each bullet point should not exceed 55 words, and the total word count of the summary should not exceed 400 words. Input call:"

[0204] If the word count of the text fragment is greater than 800 words, the corresponding second prompt word template can be:

[0205] "The following is a general type of call record with more than 800 words. Please generate a call summary according to the following template: In this call, I (the other party) (discussed, informed, communicated, consulted, requested, reminded,...) the other party (me) (discussed, informed, communicated, consulted, requested, reminded...). 1) xx 2) xx ; Please note that the number of bullet points should not exceed 6, each bullet point should not exceed 60 words, and the total number of words in the summary should not exceed 450 words. Input call:"

[0206] In the examples of the second prompt word template corresponding to five different length ranges under the above general type, the text fragment to be filled is in the second call slot after "Input call:". "\n" represents a line break.

[0207] As an example, in the scenario of a consultation type call, according to the different lengths of the call transcription text, the second prompt word template can be:

[0208] If the number of words in the text fragment is less than 81, the corresponding second prompt word template can be:

[0209] "The following is a consultation type call record with 10 - 80 words. Please generate a call summary according to the following template: In this call, (I / the other party) consulted (the other party / me) about the xx event xxx,... Please note that only one sentence needs to be generated, and the total number of words in the summary is recommended not to exceed 60% of the input call words. Input call:"

[0210] If the number of words in the text fragment is greater than 80 and less than 251, the corresponding second prompt word template can be:

[0211] "The following is a consultation type call record with 80 - 250 words. Please generate a call summary according to the following template: In this call, xx consulted about the xx event, mainly including the following: 1) Consultation about... xx's reply:... 2) Consultation about... xx's reply:... (If both parties have consulted each other) At the same time, xx consulted about the xx event, mainly including the following: 1) Consultation about... xx's reply:... 2) Consultation about... xx's reply:... ; Please note that the number of bullet points is recommended not to exceed 3, each bullet point is recommended not to exceed 50 words, and the total number of words in the summary is recommended not to exceed 125. Input call:"

[0212] If the number of words in the text fragment is greater than 250 and less than 451, the corresponding second prompt word template can be:

[0213] "The following is a consultation-type call record with a word count of 250 - 450. Please generate a call summary according to the following template: In this call, xx consulted about xx event, which mainly includes the following: \n 1) Consultation about... \n xx replied:... \n 2) Consultation about... \n xx replied:.... \n (If both parties have consulted each other) At the same time, xx consulted about xx event, which mainly includes the following: \n 1) Consultation about... \n xx replied:... \n 2) Consultation about... \n xx replied:.... ; \n Please note that the number of bullet points is recommended not to exceed 4, and each bullet point is recommended not to exceed 55 words. The total word count of the summary is recommended not to exceed 215. \n Input call: "

[0214] If the word count of the text fragment is more than 450 words and less than 801 words, the corresponding second prompt word template can be:

[0215] "The following is a consultation-type call record with a word count of 450 - 800. Please generate a call summary according to the following template: In this call, xx consulted about xx event, which mainly includes the following: \n 1) Consultation about... \n xx replied:... \n 2) Consultation about... \n xx replied:.... \n (If both parties have consulted each other) At the same time, xx consulted about xx event, which mainly includes the following: \n 1) Consultation about... \n xx replied:... \n 2) Consultation about... \n xx replied:.... ; \n Please note that the number of bullet points is recommended not to exceed 5, and each bullet point is recommended not to exceed 60 words. The total word count of the summary is recommended not to exceed 350. \n Input call: "

[0216] If the word count of the text fragment is more than 800 words, the corresponding second prompt word template can be:

[0217] "The following is a consultation-type call record with a word count of over 800. Please generate a call summary according to the following template: In this call, xx consulted about xx event, which mainly includes the following: \n 1) Consultation about... \n xx replied:... \n 2) Consultation about... \n xx replied:.... \n (If both parties have consulted each other) At the same time, xx consulted about xx event, which mainly includes the following: \n 1) Consultation about... \n xx replied:... \n 2) Consultation about... \n xx replied:.... ; \n Please note that the number of bullet points is recommended not to exceed 6, and each bullet point is recommended not to exceed 60 words. The total word count of the summary is recommended not to exceed 400. \n Input call: "

[0218] In the examples of the second prompt word templates corresponding to five different length ranges under the above consultation type, after "Input call:", there is a second call slot for filling text segments. "\n" indicates a line break.

[0219] As an example, in a scenario where the call type is a discussion type, according to the different lengths of the call transcription text, the second prompt word template can be:

[0220] If the number of words in the text segment is less than 81 words, the corresponding second prompt word template can be:

[0221] "The following is a discussion-type call record with the number of words between 10 and 80. Please generate a call summary according to the following template: In this call, I discussed with the other party...; please note that only one sentence needs to be generated, and the total number of words in the summary should not exceed 60% of the number of words in the input call. \nInput call:"

[0222] If the number of words in the text segment is greater than 80 words and less than 251 words, the corresponding second prompt word template can be:

[0223] "The following is a discussion-type call record with the number of words between 80 and 250. Please generate a call summary according to the following template: In this call, I discussed with the other party...\n1) (Subtitle):...\nConclusion:...\n2) (Subtitle):...\nConclusion:...\n3) (Subtitle):...\nConclusion:...; please note that the number of bullet points should not exceed 3, each bullet point should not exceed 50 words, and the total number of words in the summary should not exceed 150 words. \nInput call:"

[0224] If the number of words in the text segment is greater than 250 words and less than 451 words, the corresponding second prompt word template can be:

[0225] "The following is a discussion-type call record with the number of words between 250 and 450. Please generate a call summary according to the following template: In this call, I discussed with the other party...\n1) (Subtitle):...\nConclusion:...\n2) (Subtitle):...\nConclusion:...\n3) (Subtitle):...\nConclusion:...; please note that the number of bullet points should not exceed 4, each bullet point should not exceed 60 words, and the total number of words in the summary should not exceed 220 words. \nInput call:"

[0226] If the number of words in the text segment is greater than 450 words and less than 801 words, the corresponding second prompt word template can be:

[0227] "The following is a call record of the discussion type, with the number of words between 450 and 800. Please generate a call summary according to the following template: In this call, I discussed with the other party...\n1) (Subtitle):...\nConclusion:...\n2) (Subtitle):...\nConclusion:...\n3) (Subtitle):...\nConclusion:...; Please note that the number of bullet points should not exceed 5, each bullet point should not exceed 65 words, and the total number of words in the summary should not exceed 400 words.\nInput call: "

[0228] If the number of words in the text fragment is greater than 800, the corresponding second prompt word template can be:

[0229] "The following is a call record of the discussion type, with the number of words above 800. Please generate a call summary according to the following template: In this call, I discussed with the other party...\n1) (Subtitle):...\nConclusion:...\n2) (Subtitle):...\nConclusion:...\n3) (Subtitle):...\nConclusion:...; Please note that the number of bullet points should not exceed 7, each bullet point should not exceed 70 words, and the total number of words in the summary should not exceed 450 words.\nInput call: "

[0230] In the examples of the second prompt word templates corresponding to five different length ranges under the above discussion type, the text after "Input call:" is the second call slot for filling the text fragment. "\n" represents a line break.

[0231] A flowchart of an implementation for integrating the information of the summaries of each text fragment provided by the embodiments of the present application is as Figure 6 shown, and may include:

[0232] Step S601: Concatenate the summaries of each text fragment into one or more concatenated summaries; each concatenated summary is obtained by concatenating the summaries of at least two consecutive text fragments; different concatenated summaries cover the summaries of different text fragments.

[0233] It is possible to determine whether to concatenate the summaries of each text fragment into one concatenated summary or multiple concatenated summaries according to the total length of the summaries of each text fragment.

[0234] Optionally, if the total length of the summaries of each text fragment is less than or equal to the preset length, the summaries of each text fragment can be concatenated into one concatenated summary.

[0235] If the total length of the summaries of each text fragment is greater than the preset length, the summaries of each text fragment can be concatenated into at least two concatenated summaries. The length of each concatenated summary is less than or equal to the preset length. For example, if there are 5 text fragments in total, the summaries of the first two text fragments can be concatenated into one concatenated summary, and the summaries of the last three text fragments can be concatenated into the second concatenated summary.

[0236] Step S602: For each spliced minutes, add the spliced minutes to the minutes segment slot of the third prompt word template corresponding to the call type to obtain the third prompt word corresponding to the spliced minutes; the third prompt word template also includes the call type and the third task instruction, and the third task instruction is: sort out the spliced minutes in the minutes segment slot to obtain a complete call minutes.

[0237] As an example, the third prompt word template corresponding to any call type (for the convenience of narration, denoted as xxx type) can be:

[0238] "The following is a segment of a call minutes of the xxx type, which has been sorted in chronological order. Please sort out these minutes segments, put them together in chronological order, merge some similar content, organize some disordered numbers, and finally output a complete call minutes. Pay special attention that all content should be organized in chronological order during the merging process and no disorder is allowed. Input call minutes segments: "

[0239] In the above third prompt word template, after "Input call minutes segments: " is the minutes segment slot.

[0240] For example, the third prompt word template corresponding to the consultation type can be:

[0241] "The following is a segment of a call minutes of the consultation type, which has been sorted in chronological order. Please sort out these minutes segments, put them together in chronological order, merge some similar content, organize some disordered numbers, and finally output a complete call minutes. Pay special attention that all content should be organized in chronological order during the merging process and no disorder is allowed. Input call minutes segments: "

[0242] For another example, the third prompt word template corresponding to the discussion type can be:

[0243] "The following is a segment of a call minutes of the discussion type, which has been sorted in chronological order. Please sort out these minutes segments, put them together in chronological order, merge some similar content, organize some disordered numbers, and finally output a complete call minutes. Pay special attention that all content should be organized in chronological order during the merging process and no disorder is allowed. Input call minutes segments: "

[0244] Step S603: Input the third prompt word corresponding to the spliced minutes into the large model to obtain the integration result of the spliced minutes generated by the large model.

[0245] In the case of splicing the minutes of each text segment in chronological order to obtain a spliced minutes, the integration result of the spliced minutes is the complete call minutes in the format corresponding to the call type, that is, the complete call minutes extracted from the call transcription text.

[0246] As shown in Figure 7 , an example is provided in an embodiment of the present application for integrating the minutes of two text fragments ( Figure 7 respectively recorded as call sub-summary 1 and call sub-summary 2 in Figure 7 ) to obtain a complete call minutes in a format corresponding to the call type ( Figure 7 recorded as full-text call minutes in

[0247] In the case where the minutes of each text fragment are spliced in chronological order to obtain multiple spliced minutes, it further includes:

[0248] Step S604: After splicing the integration results of each spliced minutes, add them to the third prompt word template to obtain the third prompt word corresponding to the spliced integration result; input the third prompt word corresponding to the spliced integration result into the large model to obtain a complete call minutes in a format corresponding to the call type generated by the large model (i.e., the call minutes extracted from the call transcription text).

[0249] In the case of obtaining multiple spliced minutes, it is necessary to integrate the integration results of multiple spliced minutes again to obtain the complete call minutes extracted from the call transcription text. Based on this, the present application splices the integration results of each spliced minutes in chronological order and adds them to the third prompt word template corresponding to the call type to obtain the third prompt word corresponding to the spliced integration result; input the third prompt word corresponding to the spliced integration result into the large model to obtain the final call minutes in a format corresponding to the call type generated by the large model, that is, the complete call minutes extracted from the call transcription text.

[0250] In an optional embodiment, the information processing method of the present application can also perform to-do extraction in addition to minutes extraction. Based on this, the information processing method provided in an embodiment of the present application may further include:

[0251] Perform to-do extraction on the call transcription text to obtain the to-do items of each speaker and the retrieval sentences corresponding to each to-do item.

[0252] Optionally, the to-do items of each speaker are listed separately. As Figure 8a shown, an example diagram of the to-do items of both parties in a call provided in an embodiment of the present application. By listing the to-do items of different speakers separately (i.e., listing by speaker), it is convenient for users to quickly and clearly determine all their own to-do items and all the to-do items of the other party.

[0253] The retrieval sentence corresponding to each to-do item is the sentence in the call transcription text corresponding to that to-do item.

[0254] By extracting the retrieval sentences corresponding to the to-do items, it can help the user trace back to the time point and conversation information when the to-do items occurred in the original call, and help the user better understand and accept the to-do items.

[0255] Such as Figure 8b shown, it is an example of the extracted to-do items and retrieval sentences provided by the embodiment of the present application. In this example, there are only the to-do items of "me", and no to-do items of the "other party" who called with "me".

[0256] In an optional embodiment, a flowchart of an implementation of the above-mentioned extraction of to-do items from the call record text is as Figure 9 shown, and it may include:

[0257] Step S901: Divide the call transcription text into at least one text segment; each text segment contains consecutive content in the call transcription text; the length of each text segment is less than or equal to a preset length.

[0258] This step and Figure 2 the step S201 in the embodiment shown may be the same step.

[0259] Step S902: For each text segment, add the text segment to the third call slot of the fourth prompt word template to obtain the corresponding fourth prompt word for the text segment.

[0260] The fourth prompt word template also includes a fourth task instruction, and the fourth task instruction is: Extract the to-do items that each speaker needs to execute and the corresponding retrieval sentences for each to-do item from the text segments in the third call slot according to the specified requirements.

[0261] As an example, the fourth prompt word template may be:

[0262] "Please extract the to-do items and the corresponding retrieval sentences that <me> and <the other party> need to execute from the following call. The retrieval sentence corresponding to each to-do item is the sentence in the call corresponding to that to-do item. To-do items are the key tasks that need to take actions or adjust plans after the call ends. Generation requirements: All to-do items in the conversation must be listed to ensure no omission, each to-do item is only extracted once, and avoid extracting content that is not a to-do item. The to-do items of both parties are represented separately and are generated in the format of "My to-do: to-do item\nOther party's to-do: to-do item". In each to-do item, if the execution time is involved, it is presented in the format of "#Execution time#". If there are no to-do items in the whole call, output "No to-do". Input call: "

[0263] In the example of the above fourth prompt word template, after "Input call:", it is the third call slot.

[0264] Step S903: Input the fourth prompt word corresponding to the text segment into the large model to obtain the to-do items of each speaker corresponding to the text segment generated by the large model, and the retrieval sentences corresponding to each to-do item.

[0265] In the case where the call transcription text is divided into one text segment, the to-do items of each speaker generated by the large model and the retrieval sentences corresponding to each to-do item are the final to-do extraction results.

[0266] Step S904: In the case where the call transcription text is divided into at least two text segments, merge the to-do items and corresponding retrieval sentences of the same speaker corresponding to each text segment to obtain the to-do items of each speaker corresponding to the call transcription text, and the retrieval sentences corresponding to each to-do item.

[0267] Merging the to-do items of the same speaker means removing duplicates and sorting the to-do items of the same speaker corresponding to different text segments.

[0268] As Figure 10 shown, it is an example of merging the to-do items of two text segments ( Figure 10 denoted as call sub-to-do 1 and call sub-to-do 2 respectively in Figure 10 ) provided by the embodiment of the present application to obtain the final to-do item ( Figure 10 denoted as full text to-do in

[0269] In the full text to-do, the retrieval sentence of each to-do item can be displayed below the to-do item.

[0270] To improve the to-do recall rate of the large model of the present application (denoted as the first large model for easy description and distinction), the training data set can be collected in the following way, and the first large model can be further trained based on the collected training data set to improve the to-do recall rate of the large model.

[0271] Collect several call transcription texts as several training samples, that is, each training sample is a call transcription text.

[0272] For each training sample, add the training sample to the third call slot of the fourth prompt word template to obtain the fourth prompt word corresponding to the training sample.

[0273] Input the fourth prompt word corresponding to the training sample into the first large model to obtain a first extraction result; the first extraction result represents the to-do items of each speaker extracted by the first large model from the training sample and the retrieval sentences corresponding to the to-do items.

[0274] If the first extraction result represents that to-do items are extracted from the training sample, input the fourth prompt word corresponding to the training sample into the second large model twice to obtain a second extraction result and a third extraction result respectively; the second extraction result represents the to-do items of each speaker extracted by the second large model from the training sample for the first time and the retrieval sentences corresponding to the to-do items; the third extraction result represents the to-do items of each speaker extracted by the second large model from the training sample for the second time and the retrieval sentences corresponding to the to-do items. The to-do recall rate of the second large model is greater than that of the first large model. That is to say, in this application, the second large model is called twice, and each time the second large model is called, the training sample is input into the second large model.

[0275] For each to-do item of each speaker in the second extraction result, calculate the similarity between this to-do item and each to-do item of the same speaker in the third extraction result (for example, for each to-do item of "I" in the second extraction result, calculate the similarity between this to-do item and each to-do item of "I" in the third extraction result; for each to-do item of "the other party" in the second extraction result, calculate the similarity between this to-do item and each to-do item of "the other party" in the third extraction result). If there is a to-do item in the third extraction result whose similarity to this to-do item is greater than the similarity threshold, retain this to-do item in the second extraction result as a candidate to-do item; otherwise, filter out this to-do item in the second extraction result.

[0276] Clean the candidate to-do items to filter out the candidate to-do items that are inconsistent with the original text semantics. Optionally, for any candidate to-do item, the person name and time in this candidate to-do item can be extracted, and the person name and time in the text segment corresponding to this candidate to-do item can be extracted. If the person name extracted from this candidate to-do item is the same as the person name extracted from the text segment, and the time extracted from this candidate to-do item is within the range of the time extracted from the text segment, retain this candidate to-do item; otherwise, filter out this candidate to-do item.

[0277] Mark the retained candidate to-do items and their retrieval sentences, as well as the to-do items and their retrieval sentences in the first extraction result, as the labels of the training sample.

[0278] If the first extraction result represents that there are no to-do items in the training sample, process the next training sample.

[0279] Based on this embodiment, a large number of training samples can be quickly obtained, improving the model training efficiency and to-do recall rate.

[0280] Corresponding to the method embodiments, the present application also provides an information processing device. A schematic structural diagram of the information processing device provided in the embodiments of the present application is shown as Figure 11 follows and may include:

[0281] An identification module 1101, a type determination module 1102, and a summary extraction module 1103;

[0282] Among them, the identification module 1101 is used to perform speech recognition on the call voice to obtain a call transcription text;

[0283] The type determination module 1102 is used to determine the call type based on the call transcription text; different call types correspond to different call purposes;

[0284] The summary extraction module 1103 is used to extract a summary from the call transcription text based on the call type to obtain a call summary in a format corresponding to the call type.

[0285] The information processing device provided in the embodiments of the present application performs speech recognition on the call voice to obtain a call transcription text, determines the call type representing the call purpose based on the call transcription text, and extracts a summary from the call transcription text based on the call type to obtain a call summary in a format corresponding to the call type. Based on the present application, different formats of call summaries can be obtained according to different call types, improving the diversity of the generated call summary formats, thereby enhancing the auxiliary effect of the call summary on users.

[0286] In an optional embodiment, when the type determination module 1102 determines the call type based on the call transcription text, it is used to:

[0287] Divide the call transcription text into at least one text segment; each text segment contains consecutive content in the call transcription text; the length of each text segment is less than or equal to a preset length;

[0288] Determine the call type based on at least one text segment;

[0289] Among them, determining the call type based on one text segment includes: adding the one text segment to the first call slot of the first prompt word template to obtain a first prompt word; the first prompt word template further includes a first task instruction, and the first task instruction is: determine the call type based on the text segment in the first call slot; input the first prompt word into a large model to obtain the call type output by the large model based on the text segment in the first call slot;

[0290] Determining the call type based on at least two text segments includes: for each text segment, adding the text segment to the first call slot of the first prompt word template to obtain the first prompt word corresponding to the text segment; inputting the first prompt word corresponding to the text segment into the large model to obtain the call type corresponding to the text segment output by the large model based on the text segment in the first call slot; determining the final call type based on the call types corresponding to each text segment in the at least two text segments.

[0291] In an optional embodiment, when the type determination module 1102 determines the final call type based on the call types corresponding to each text segment in the at least two text segments, it is configured to:

[0292] If the call types corresponding to each text segment in the at least two text segments are the same, determining the call type corresponding to any one of the text segments as the final call type;

[0293] If there are at least two text segments in the at least two text segments whose corresponding call types are different, determining the general type as the final call type.

[0294] In an optional embodiment, the speaker information and its corresponding speech content are recorded in the call transcription text in the chronological order of the speaker's speech; each speaker information and its corresponding speech content are marked as a paragraph; when the type determination module 1102 divides the call transcription text into at least two text segments:

[0295] Each text segment includes one or consecutive multiple paragraphs in the call transcription text;

[0296] Two adjacent text segments do not cover the same paragraph, and two adjacent text segments cover consecutive multiple paragraphs in the call transcription text.

[0297] In an optional embodiment, when the summary extraction module 1103 extracts a summary from the call transcription text based on the call type, it is configured to:

[0298] For each text segment, adding the text segment to the second call slot of the second prompt word template corresponding to the call type to obtain the second prompt word corresponding to the text segment; the second prompt word template further includes a second task instruction, the call type, and its corresponding summary template, and the second task instruction is: generating a summary of the text segment in the second call slot according to the summary template; different summary templates correspond to different summary formats;

[0299] Inputting the second prompt word corresponding to the text segment into the large model to obtain the summary of the text segment generated by the large model according to the summary template;

[0300] In the case of dividing the call transcription text into at least two text segments, integrate the information of the minutes of each text segment to obtain a complete call minutes in a format corresponding to the call type.

[0301] In an optional embodiment, for each text segment, when the minutes extraction module 1103 adds the text segment to the second call slot of the second prompt word template corresponding to the call type, it is used for:

[0302] Determine the length range to which the length of the text segment belongs;

[0303] Add the text segment to the second call slot of the second prompt word template corresponding to the call type and the length range; for different length ranges, the minutes templates corresponding to the call type in the second prompt word template are different.

[0304] In an optional embodiment, when the minutes extraction module 1103 integrates the information of the minutes of each text segment to obtain a complete call minutes in a format corresponding to the call type, it is used for:

[0305] Concatenate the minutes of each text segment into one or more concatenated minutes; each concatenated minute is obtained by concatenating the minutes of at least two consecutive text segments; different concatenated minutes cover the minutes of different text segments;

[0306] For each concatenated minute, add the concatenated minute to the minutes segment slot of the third prompt word template corresponding to the call type to obtain the third prompt word corresponding to the concatenated minute; the third prompt word template also includes the call type and a third task instruction, and the third task instruction is: sort out the concatenated minutes in the minutes segment slot to obtain a complete call minutes;

[0307] Input the third prompt word corresponding to the concatenated minute into the large model to obtain the integration result of the concatenated minute generated by the large model;

[0308] In the case of concatenating the minutes of each text segment into one concatenated minute, the integration result of the concatenated minute is a complete call minutes in a format corresponding to the call type;

[0309] In the case of concatenating the minutes of each text segment into multiple concatenated minutes, concatenate the integration results of each concatenated minute and add them to the third prompt word template to obtain the third prompt word corresponding to the concatenated integration result; input the third prompt word corresponding to the concatenated integration result into the large model to obtain a complete call minutes in a format corresponding to the call type generated by the large model.

[0310] In an optional embodiment, the information processing device further includes a to-do extraction module, which is used for:

[0311] Performing to-do extraction on the call transcription text to obtain to-do items of each speaker and retrieval sentences corresponding to each to-do item;

[0312] The retrieval sentence corresponding to each to-do item is the sentence in the call transcription text corresponding to the to-do item.

[0313] In an optional embodiment, when the to-do extraction module performs to-do extraction on the call transcription text, it is used for:

[0314] Dividing the call transcription text into at least one text segment; each text segment contains consecutive content in the call transcription text; the length of each text segment is less than or equal to a preset length;

[0315] For each text segment, adding the text segment to the third call slot of the fourth prompt template to obtain the fourth prompt corresponding to the text segment; the fourth prompt template further includes a fourth task instruction, and the fourth task instruction is: extracting to-do items that each speaker needs to execute and retrieval sentences corresponding to each to-do item from the text segment in the third call slot according to specified requirements;

[0316] Inputting the fourth prompt corresponding to the text segment into a large model to obtain to-do items of each speaker corresponding to the text segment generated by the large model and retrieval sentences corresponding to each to-do item;

[0317] In the case of dividing the call transcription text into at least two text segments, merging the to-do items and corresponding retrieval sentences of the same speaker corresponding to each text segment to obtain to-do items of each speaker corresponding to the call transcription text and retrieval sentences corresponding to each to-do item.

[0318] An electronic device is also provided in an embodiment of the present application. Refer to Figure 12 As shown, it shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application can be a terminal device (such as a car machine, a large screen device, a smart home, a mobile phone, a tablet computer, a notebook computer, a desktop computer, etc.), or a server (which can be a single server, a server cluster, or a cloud server, etc.). Figure 12 The electronic device shown is only an example and should not bring any limitation to the functions and usage scope of the embodiments of the present application.

[0319] As Figure 12As shown, the electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 1201, which may perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1202 or the program loaded from the storage device 1208 into the random access memory (RAM) 1203. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 1203. The processing device 1201, the ROM 1202, and the RAM 1203 are connected to each other through a bus 1204. The input / output (I / O) interface 1205 is also connected to the bus 1204.

[0320] Generally, the following devices may be connected to the I / O interface 1205: an input device 1206 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 1207 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1208 including, for example, a memory card, a hard disk, etc.; and a communication device 1209. The communication device 1209 may allow the electronic device to communicate with other devices wirelessly or wiredly to exchange data. Although FIG. 8 shows an electronic device having various devices, it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices may be implemented or had.

[0321] An embodiment of the present application also provides a computer program product including computer-readable instructions, which, when running on an electronic device, enable the electronic device to implement any of the information processing methods provided by the embodiments of the present application.

[0322] An embodiment of the present application also provides a computer-readable storage medium carrying one or more computer programs, which, when executed by an electronic device, can enable the electronic device to implement any of the information processing methods provided by the embodiments of the present application.

[0323] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationships between the modules indicate that they have communication connections, which may be specifically implemented as one or more communication buses or signal lines.

[0324] Through the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general hardware. Of course, it can also be implemented by dedicated hardware including application-specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. Generally, functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures for implementing the same function can also be various, such as analog circuits, digital circuits, or dedicated circuits, etc. However, for this application, in more cases, software program implementation is a better implementation method. Based on such an understanding, the technical solution of this application, in essence, or the part that makes a contribution to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disc of a computer, etc., and includes several instructions to enable a computer device (which can be a personal computer, training device, or network device, etc.) to execute the methods described in various embodiments of this application.

[0325] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. Professional technicians can use different methods to implement the described functions for each specific solution, but such implementation should not be considered to exceed the scope of this application.

[0326] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are generated in whole or in part. The computer can be a general-purpose computer, a dedicated computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, training device, or data center to another website, computer, training device, or data center by wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can store, or a data storage device such as a training device or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0327] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference may be made to each other.

[0328] 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 obvious to those skilled in the art. The general principles defined herein can 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 rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An information processing method, characterized in that: include: Perform speech recognition on the call voice to obtain the call transcription text; determining a call type based on the call transcript; Different call types correspond to different call purposes; Minutes are extracted from the call transcription text based on the call type to obtain call minutes in a format corresponding to the call type.

2. The method according to claim 1, characterized in that The determining the call type based on the call transcription text includes: Dividing the call transcription text into at least one text segment; each text segment contains continuous content in the call transcription text; the length of each text segment is less than or equal to a preset length; determining a call type based on at least one text segment; Wherein, determining the call type based on a text segment includes: adding the text segment to the first call slot of the first prompt word template to obtain a first prompt word; the first prompt word template also includes a first task instruction, and the first task instruction is: judging the call type based on the text segment in the first call slot; inputting the first prompt word into the large model to obtain the call type output by the large model based on the text segment in the first call slot; Determining the call type based on at least two text fragments includes: for each text fragment, adding the text fragment to the first call slot of the first prompt word template to obtain the first prompt word corresponding to the text fragment; inputting the first prompt word corresponding to the text fragment into the large model to obtain the call type corresponding to the text fragment output by the large model based on the text fragment in the first call slot; and determining the final call type based on the call types corresponding to each of the at least two text fragments.

3. The method according to claim 2, characterized in that The determining the final call type based on the call type corresponding to each of the at least two text segments includes: If the call types corresponding to the respective text segments in the at least two text segments are the same, determining the call type corresponding to any one of the text segments as the final call type; If at least two of the at least two text segments correspond to different call types, the common type is determined as the final call type.

4. The method according to claim 2, characterized in that: The call transcription text records the speaker information and the corresponding voice content in the order of the speaker's speaking time; each speaker information and the corresponding voice content are marked as a paragraph; when the call transcription text is divided into at least two text segments: Each text segment includes one or more consecutive paragraphs in the call transcription text; Two adjacent text segments do not cover the same paragraph, and two adjacent text segments cover multiple consecutive paragraphs in the call transcription text.

5. The method according to claim 2, characterized in that: The extracting a summary of the call transcription text based on the call type includes: For each text segment, add the text segment to the second call slot of the second prompt word template corresponding to the call type to obtain the second prompt word corresponding to the text segment; the second prompt word template also includes a second task instruction, the call type and its corresponding minutes template, and the second task instruction is: generate minutes of the text segment in the second call slot according to the minutes template; different minutes templates correspond to different minutes formats; Inputting the second prompt word corresponding to the text segment into the macro model to obtain the minutes of the text segment generated by the macro model according to the minutes template; When the call transcription text is divided into at least two text segments, the minutes of each text segment are integrated to obtain a complete call minute in a format corresponding to the call type.

6. The method according to claim 5, characterized in that For each text segment, adding the text segment to the second conversation slot of the second prompt word template corresponding to the conversation type includes: Determine the length range to which the length of the text segment belongs; The text segment is added to the second call slot of the second prompt word template corresponding to the call type and the length range; corresponding to different length ranges, the minutes template corresponding to the call type in the second prompt word template is different.

7. The method according to claim 5, characterized in that The minutes of each text segment are integrated to obtain a complete call minutes in a format corresponding to the call type, including: The minutes of each text segment are spliced ​​into one or more spliced ​​minutes; each spliced ​​minute is obtained by splicing the minutes of at least two consecutive text segments; different spliced ​​minutes cover the minutes of different text segments; For each spliced ​​minutes, add the spliced ​​minutes to the minutes segment slot of the third prompt word template corresponding to the call type to obtain the third prompt word corresponding to the spliced ​​minutes; the third prompt word template also includes the call type and the third task instruction, and the third task instruction is: sort out the spliced ​​minutes in the minutes segment slot to obtain a complete call minutes; Inputting the third prompt word corresponding to the spliced ​​minutes into the large model to obtain the integration result of the spliced ​​minutes generated by the large model; In the case of splicing the minutes of the various text segments into a spliced ​​minute, the integration result of the spliced ​​minute is a complete call minute in a format corresponding to the call type; When the minutes of each text fragment are spliced ​​into multiple spliced ​​minutes, the integration results of each spliced ​​minute are spliced ​​and added to the third prompt word template to obtain the third prompt word corresponding to the spliced ​​integration result; the third prompt word corresponding to the spliced ​​integration result is input into the big model to obtain a complete call minutes generated by the big model in a format corresponding to the call type.

8. The method according to any one of claims 1 to 7, characterized in that: Also includes: Performing to-do extraction on the call transcription text to obtain to-do items of each speaker and search sentences corresponding to each to-do item; The search sentence corresponding to each to-do item is a sentence in the call transcription text corresponding to the to-do item.

9. The method according to claim 8, characterized in that Performing to-do extraction on the call transcription text includes: Dividing the call transcription text into at least one text segment; each text segment contains continuous content in the call transcription text; the length of each text segment is less than or equal to a preset length; For each text segment, add the text segment to the third call slot of the fourth prompt word template to obtain a fourth prompt word corresponding to the text segment; the fourth prompt word template also includes a fourth task instruction, and the fourth task instruction is: extract the to-do items to be executed by each speaker and the search sentence corresponding to each to-do item from the text segment in the third call slot according to the specified requirements; Inputting the fourth prompt word corresponding to the text segment into the large model, obtaining the to-do items of each speaker corresponding to the text segment generated by the large model, and the search sentences corresponding to each to-do item; When the call transcription text is divided into at least two text segments, the to-do items and corresponding search sentences of the same speaker corresponding to each text segment are merged to obtain the to-do items of each speaker corresponding to the call transcription text and the search sentences corresponding to each to-do item.

10. An information processing device, characterized in that: include: A recognition module is used to perform speech recognition on the call speech and obtain the call transcription text; A type determination module, configured to determine a call type based on the call transcription text; Different call types correspond to different call purposes; The minutes extraction module is used to extract minutes from the call transcription text based on the call type to obtain a call minutes in a format corresponding to the call type.

11. A computer program product, characterized in that The method comprises computer-readable instructions, and when the computer-readable instructions are executed on an electronic device, the electronic device implements the information processing method according to any one of claims 1 to 9.

12. An electronic device, characterized in that: The electronic device comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the information processing method according to any one of claims 1 to 9.

13. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the information processing method as described in any one of claims 1 to 9.

Images