Data processing method, device and equipment, computer readable storage medium and product

Abstract

The invention provides a data processing method, device and equipment, a computer readable storage medium and a product, and relates to the field of artificial intelligence, in particular to the field of large language models and intelligent questions and answers. The specific implementation scheme is as follows: obtaining text content published by a user for any target object associated with live broadcast content in a live broadcast room, and identifying a target intention associated with the text content through a preset intention classification model; adopting a network model matched with the target intention to generate a reply text corresponding to the text content; generating target reply content according to associated data of the user, the reply text and parameter information of the target object, wherein the associated data comprises identification information of the user and the text content; and performing a reply operation on the text content based on the target reply content within a reply opportunity meeting a preset condition. Therefore, on the basis of not interrupting the live broadcast process, accurate reply operation can be performed on the text content issued by the user.

Description

Technical Field

[0001] The present disclosure relates to large language models and intelligent question and answer in artificial intelligence, and in particular to a data processing method, device, equipment, computer-readable storage medium and product. Background Art

[0002] During the live broadcast, the host can explain the target object. While watching the live broadcast, users can make comments in real time in the live broadcast room, where the comments can be questions about the target object and chat content. The host can reply to the comments posted by the user to answer the user's comments on the target object.

[0003] In the related art, in order to improve the efficiency of live broadcasting, a virtual digital person can be set as the anchor of the live broadcast room to perform live broadcasting operations. How to make the virtual digital person accurately reply to the comments posted by users has become an urgent problem to be solved. Summary of the invention

[0004] The present disclosure provides a data processing method, device, equipment, computer-readable storage medium and product for improving the accuracy and real-time performance of intelligent questions and answers in a live broadcast room.

[0005] According to a first aspect of the present disclosure, there is provided a data processing method, comprising:

[0006] Acquire text content posted by a user in the live broadcast room for any target object associated with the live broadcast content, and identify the target intent associated with the text content through a preset intent classification model;

[0007] Using a network model that matches the target intent to generate a reply text corresponding to the text content;

[0008] Generate target reply content according to the user's associated data, the reply text and the parameter information of the target object, wherein the associated data includes the user's identification information and the text content;

[0009] A reply operation is performed on the text content based on the target reply content within a reply opportunity that meets preset conditions.

[0010] According to a second aspect of the present disclosure, there is provided a data processing device, comprising:

[0011] An acquisition module is used to acquire text content published by a user in a live broadcast room for any target object associated with the live broadcast content, and identify the target intent associated with the text content through a preset intent classification model;

[0012] A generation module, used to generate a reply text corresponding to the text content by using a network model matching the target intent;

[0013] A processing module, configured to generate target reply content according to the associated data of the user, the reply text and parameter information of the target object, wherein the associated data includes the identification information of the user and the text content;

[0014] The reply module is used to reply to the text content based on the target reply content within a reply opportunity that meets preset conditions.

[0015] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0019] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method described in the first aspect.

[0020] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising: a computer program, wherein the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program so that the electronic device executes the method described in the first aspect.

[0021] The technology disclosed in the present invention improves the accuracy of the virtual anchor's reply content during the live broadcast process without interrupting the live broadcast process, thereby improving the live broadcast effect.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0024] Figure 1 A system architecture diagram based on the embodiments of the present disclosure;

[0025] Figure 2 A flowchart of a data processing method provided by an embodiment of the present disclosure;

[0026] Figure 3 A schematic diagram of an application scenario provided for an embodiment of the present disclosure;

[0027] Figure 4 An architectural diagram of target reply content provided for an embodiment of the present disclosure;

[0028] Figure 5 A flowchart of a data processing method provided by another embodiment of the present disclosure;

[0029] Figure 6 A schematic diagram of the structure of a data processing device provided in an embodiment of the present disclosure;

[0030] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0032] The present disclosure provides a data processing method, device, equipment, computer-readable storage medium and product, which are applied to large language models and intelligent question and answer in the field of artificial intelligence, so as to achieve the technical effect of improving the accuracy of the virtual anchor's replies during the live broadcast process.

[0033] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0034] During the live broadcast, the host can explain the target object. While watching the live broadcast, users can make comments in real time in the live broadcast room, where the comments can be questions about the target object and chat content. The host can reply to the comments posted by the user to answer the user's comments on the target object.

[0035] In the related art, a question-and-answer database can be pre-built, and after obtaining the comments posted by the user, the reply content matching the comments can be determined in the question-and-answer database. A reply voice is generated based on the reply content to control the virtual anchor to reply based on the reply voice.

[0036] However, since the question and answer content stored in the question and answer database is limited, the answer content generated by the above method often cannot cover all the questions raised by the user, and the answer content that may be generated cannot meet the actual needs of the user.

[0037] In the process of solving the above technical problems, the inventors discovered through research that in order to improve the accuracy of intelligent questions and answers of virtual anchors during live broadcasts, the reply text can be generated by combining the intent classification model and the big model.

[0038] In addition, in order to improve the user experience of the live broadcast room audience, after generating the reply text, the audience's name and question can be repeated to strengthen the user's impression of the reply text. In addition, parameter information associated with the target object can be added during the reply process to enable the audience to further understand the target object.

[0039] Furthermore, the timing of replying during the live broadcast can be determined, so that comments posted by viewers can be replied to in a timely manner without affecting the live broadcast process, thereby improving the content quality of the live broadcast and the user experience.

[0040] In order to enable readers to more deeply understand the implementation principle of the present disclosure, the following Figure 1-Figure 7 The embodiments of the present disclosure are further refined.

[0041] Figure 1 This is a system architecture diagram based on the embodiments of the present disclosure. Figure 1 As shown, the system architecture based on the present disclosure includes at least a terminal device 11 and a server 12. The terminal device 11 is in communication connection with the server 12. A data processing device is provided in the server 12. The data processing device can be written in languages ​​such as C / C++, Java, Shell or Python; the terminal device 11 can be, for example, a mobile phone, a tablet computer, etc.

[0042] Based on the above system architecture, users can watch live broadcasts on terminal devices 11, and during the viewing process, they can post comments in the live broadcast room. Server 12 can generate reply text based on a combination of a preset intent classification model and a large language model. Target reply content is generated based on the user's associated data, the reply text, and the parameter information of the target object. Comments are replied to according to the target reply content at the appropriate reply time.

[0043] Figure 2 A flow chart of a data processing method provided by an embodiment of the present disclosure is shown in FIG. Figure 2 As shown, the method includes:

[0044] Step 201: obtain text content posted by a user in a live broadcast room for any target object associated with the live broadcast content, and identify the target intent associated with the text content through a preset intent classification model.

[0045] The execution subject of this embodiment is a data processing device. The data processing device can be coupled to the server, so that after obtaining the text content posted by the user in the live broadcast room, a reply text can be generated based on the intention classification model and the large language model, and a reply operation can be performed based on the reply text at an appropriate time.

[0046] In this embodiment, the user can watch the live broadcast content in the live broadcast room, wherein the live broadcast room can be a live broadcast room hosted by a virtual anchor. During the live broadcast process, the virtual anchor can perform live broadcast operations based on a preset live broadcast script, and can explain the target object so that the user can understand the target object more intuitively.

[0047] For example, the live broadcast room can be a live broadcast room for selling goods. During the live broadcast, the virtual anchor can explain multiple products.

[0048] Furthermore, when users watch live broadcasts in the live broadcast room, they can post text content in the comment area according to actual needs. Therefore, in order to implement the reply operation to the text content, the text content posted by the user in the live broadcast room for any target object associated with the live broadcast content can be obtained. Among them, the text content can be the content published when the virtual anchor explains the target object, or it can be the content published specifically for a target object that has been explained.

[0049] Optionally, the text content may be questions posted for the target object, for example, the user may ask the virtual host about the details and parameters of the target object currently being explained. Alternatively, the text content may be chat content, for example, the text content may be content unrelated to the target object currently being explained by the host, such as "The weather is really nice today".

[0050] Furthermore, in order to more accurately reply to the text content, the target intent corresponding to the text content can also be identified. A preset intent classification model can be used to implement the recognition operation of the target intent.

[0051] Step 202: Use a network model that matches the target intent to generate a reply text corresponding to the text content.

[0052] In this embodiment, in order to improve the fit between the reply text and the text content and improve the accuracy of the reply text, different network models can be pre-set for different intentions. The network model can be a large language model or a small model associated with the large language model.

[0053] Optionally, after determining the target intent, a network model matching the target intent may be determined, and a corresponding reply text may be generated based on the network model and the text content.

[0054] Step 203: Generate target reply content according to the user's associated data, the reply text, and the parameter information of the target object, wherein the associated data includes the user's identification information and the text content.

[0055] In this embodiment, since the virtual anchor needs to explain the target object and respond to text content posted by different users during the live broadcast, in order to arouse the user's understanding of the virtual anchor's response to the text content posted by themselves, and to understand the target object more intuitively, the target reply content can be generated based on the user's associated data, reply text and parameter information of the target object, wherein the associated data includes the user's identification information and text content, and the parameter information includes but is not limited to the price, inventory, details and other parameters of the target object.

[0056] Step 204: Perform a reply operation on the text content based on the target reply content within a reply opportunity that meets preset conditions.

[0057] In this embodiment, during the live broadcast, when the virtual anchor is explaining the target object, inserting the target reply content often interrupts the current explanation and affects the live broadcast effect. Therefore, in order to accurately reply to the text content without affecting the live broadcast process, a reply timing that meets the preset conditions can also be determined. The text content is replied based on the target reply content within the reply timing that meets the preset conditions.

[0058] The data processing method provided in this embodiment determines the reply text by combining the intent classification model and the large language model, thereby improving the matching degree between the reply text and the text content. In addition, by generating the target reply content based on the user's associated data, the reply text and the parameter information of the target object, it is possible to increase the user's attention to the target reply content on the basis of generating a reply that fits the text content, and enable the user to further understand the target object. And by determining the reply timing, it is possible to reply to the text content within the reply timing that meets the preset conditions, and it is possible to accurately reply to the text content without affecting the live broadcast process.

[0059] Further, based on any of the above embodiments, the method further includes:

[0060] The text content and the target reply content generated based on the text content are stored in association to obtain an optimized data pair.

[0061] The network model is optimized based on the optimization data within a preset time range.

[0062] In this embodiment, after replying to the text content based on the target reply content, the target reply content may be scored. The scoring operation may be implemented through a preset network model, or may be implemented manually.

[0063] Furthermore, for the target reply content whose score reaches the preset condition, the text content and the target reply content can be stored in association to obtain an optimized data pair. Therefore, the optimization operation of the large language model can be implemented based on the optimized data pair in the future.

[0064] Optionally, in order to ensure the amount of data used for the optimization operation, the optimization operation may be performed on the large language model based on the optimization data within a preset time range.

[0065] The data processing method provided in this embodiment can improve the data processing accuracy of the large language model by optimizing the large language model based on the optimized data within a preset time range, so as to generate a more accurate reply text.

[0066] Optionally, based on any of the above embodiments, before step 201, the method further includes:

[0067] A preset training data set is obtained, wherein the training data set includes a plurality of training data pairs, the training data pairs include training texts and annotation information associated with the training texts, and the annotation information includes intent information corresponding to the training texts.

[0068] The preset classification model is iteratively trained using the training data set until the classification model meets the preset convergence condition, thereby obtaining the intent classification model.

[0069] In this embodiment, in order to determine the target intent associated with the text content based on the intent classification model, it is first necessary to train the classification model to obtain the intent classification model.

[0070] Optionally, a preset training data set may be obtained, wherein the training data set includes a plurality of training data pairs, the training data pairs include training texts and annotation information associated with the training texts, and the annotation information includes intent information corresponding to the training texts. The intent information includes, but is not limited to, parameter query intent, descriptive information consultation intent, and conversation intent.

[0071] Furthermore, after obtaining the training data set, the classification model can be iteratively trained based on the training data set until the classification model meets a preset convergence condition, thereby obtaining an intent classification model.

[0072] Optionally, the training text in the training data pair can be input into the classification model to obtain the classification result output by the classification model. The loss value of the classification model is determined based on the classification result and the annotation information. Determine whether the classification model meets the convergence condition based on the loss value. Among them, the convergence condition includes but is not limited to the loss value being lower than a preset threshold, or the difference between the loss values ​​of two rounds of training is less than a preset difference threshold. Alternatively, it can be determined whether the classification model converges based on the number of training rounds and the training duration. For example, the convergence condition includes but is not limited to the number of training rounds reaching a preset round number threshold, or the training duration reaching a preset duration threshold, etc.

[0073] Furthermore, if it is determined that the classification model meets the convergence condition, the current classification model can be determined as the intent classification model. Otherwise, the parameters of the classification model can be adjusted based on the loss value until the classification model meets the preset convergence condition to obtain the intent classification model.

[0074] The data processing method provided in this embodiment obtains a training data set that includes training text and intent information associated with the training text as annotation information, so that a preset classification model can be iteratively trained based on the training data set, so that the intent classification model that reaches the preset convergence condition can accurately identify the target intent of the text content.

[0075] Further, based on any of the above embodiments, the network model includes any one of a large language model and at least one small model associated with the large language model. Before step 201, it also includes:

[0076] Preset text description information is input into the large language model, the text description information including parameter information of at least one target object associated with the live content, a live script, a preset question and answer database, and condition information associated with the reply text, and the text description information is used to guide the large language model to generate a reply text that meets the condition information based on the parameter information, the live script, and the preset question and answer database.

[0077] In this embodiment, the network model includes any one of a large language model and at least one small model associated with the large language model.

[0078] In order to be able to generate reply text based on the network model, preset text description information can be input into the large language model in advance. The text description information includes parameter information of at least one target object associated with the live content, the live script, a preset question and answer database, and condition information associated with the reply text. The text description information is used to guide the large language model to generate a reply text that meets the condition information based on the parameter information, the live script, and the preset question and answer database.

[0079] Among them, the conditional information can limit the number of words, source, symbols and other parameters of the reply text.

[0080] Taking a practical application as an example, the text description information may be:

[0081] Reference Information:

[0082] ”'

[0083] [Basic information of target object]--Information of target object

[0084] [Reference Information]—Live Script

[0085] [Frequently Asked Questions and Answers]--Customer Frequently Asked Questions

[0086] [Shopping Service]--E-commerce General Knowledge

[0087] ”'

[0088] This is a parameter query / description information consultation / conversation intention task, answering user questions as a host. The specific requirements are as follows:

[0089] 1. If you can find the answer to the question from the reference information, you need to combine the answer and directly make a corresponding reply. The answer should be controlled within 50 words, and the task is completed;

[0090] 2. If the answer to the question cannot be found in the reference information, the answer must be positive and not violate public perception. The part of the answer related to the target object must come from the reference information and should not conflict with the reference information. Users can be appropriately guided to consult customer service or view the target object details page. The word count should be controlled within 50 words.

[0091] Note: Avoid symbols other than “,.!?” in reply.

[0092] Note: Avoid markdown, line breaks, and list symbols in your reply.

[0093] Note: You must not answer questions that are not mentioned in the reference information. Do not fabricate the target object’s price, target object’s information attributes, target object’s discounts, and target object’s services.

[0094] Note: If it involves the target object's reply, the content can be divided into three parts: 1) Repeat the user's question and add the user's nickname 2) Reply to the user's question 3) Introduce the target object's current price parameters, inventory parameters, etc.

[0095] By inputting text description information into the large prediction model, the large language model can generate a response text that is more in line with actual needs based on the text description information.

[0096] The data processing method provided in this embodiment can guide the large language model to generate a reply text that is more consistent with the text content by inputting text description information into the large language model in advance.

[0097] Optionally, based on any of the above embodiments, the target intent includes parameter query intent, description information consultation intent and conversation intent associated with the target object. Step 202 includes:

[0098] If the target intent is a conversation intent, the text content is input into a preset large language model to obtain a reply text output by the large prediction model.

[0099] If the target intent is a parameter query intent, the text content is input into the small query model associated with the large language model to obtain the reply text output by the small query model.

[0100] If the target intent is to describe information consultation intent, the text content is input into the small text generation model associated with the large language model to obtain the reply text output by the small text generation model.

[0101] In this embodiment, the target intent includes parameter query intent, description information consultation intent, and conversation intent associated with the target object. For each intent, a network model can be pre-set. The network model includes any one of a large language model and at least one small model associated with the large language model.

[0102] Optionally, if the target intent is a conversation intent, the text content is input into a preset large language model to obtain the reply text output by the large prediction model. If the target intent is a parameter query intent, the text content is input into a query small model associated with the large language model to obtain the reply text output by the query small model. If the target intent is a descriptive information consultation intent, the text content is input into a text generation small model associated with the large language model to obtain the reply text output by the text generation small model.

[0103] In order to enable the network model to generate a reply text that matches the target intent, a prompt that matches the target intent can be used in advance to guide the network model. The prompt may include the associated information of the target intent, the associated parameters of the reply text, etc. For example, when the target intent is a parameter query intent, the prompt may be that if you can find the answer to the question from the reference information, you need to combine the answer and directly make a corresponding reply. The answer should be within 50 words, and the task is completed. Prior to this, you can also feed the target model some basic information of the target object, reference information, common questions and replies, and e-commerce general knowledge.

[0104] Figure 3 A schematic diagram of an application scenario provided by an embodiment of the present disclosure, such as Figure 3 As shown, after obtaining the text content 31 posted by the user, the target intent associated with the text content 31 can be identified through the preset intent classification model 32. When the target intent is a conversation intent, the text content 31 is input into the preset large language model 33 to obtain the reply text output by the large prediction model. If the target intent is a parameter query intent, the text content 31 is input into the query small model 34 associated with the large language model to obtain the reply text output by the query small model. If the target intent is to describe information consultation intent, the text content 31 is input into the text generation small model 35 associated with the large language model to obtain the reply text output by the text generation small model.

[0105] The data processing method provided in this embodiment pre-sets multiple network models that match different intent types, so that the response text can be generated based on the target intent selection and the corresponding network model, thereby improving the fit between the response text and the text content.

[0106] Optionally, based on any of the above embodiments, step 203 includes:

[0107] The identification information associated with the user and / or the text content is used as the initial text.

[0108] The reply text is used as the intermediate text, and the parameter information associated with the target object is used as the end text, wherein the parameter information includes the price parameter and the inventory parameter associated with the target object.

[0109] The initial text, the intermediate text and the end text are concatenated in sequence to obtain the target reply content.

[0110] In this embodiment, during the live broadcast, the virtual anchor needs to explain the target object and respond to the text content published by different users. In order to make the user understand the virtual anchor's response to the text content published by the user, the user's associated data can be introduced into the target response content. The associated data includes but is not limited to the user's associated identification information and / or text content.

[0111] In addition, in order to make the user more intuitively understand the target object associated with the text content, parameter information associated with the target object can also be introduced into the target reply content, wherein the parameter information associated with the target object includes price parameters and inventory parameters associated with the target object.

[0112] Optionally, a concatenation operation may be performed on the user's associated data, the reply text, and parameter information associated with the target object to obtain target reply content.

[0113] For example, the identification information and / or text content associated with the user can be used as the initial text. The reply text can be used as the intermediate text, and the parameter information associated with the target object can be used as the end text, where the parameter information includes the price parameter and inventory parameter associated with the target object. The initial text, the intermediate text, and the end text are concatenated in sequence to obtain the target reply content.

[0114] For example, in practical applications, the virtual host can be controlled to repeat the text content posted by the user, or read the user's nickname. Then read the reply text to solve the user's questions. Finally, the buying point, price, inventory and other content related to the text content can be introduced.

[0115] Figure 4 The architecture diagram of the target reply content provided by the embodiment of the present disclosure is as follows: Figure 4 As shown, the target reply content 41 may be composed of the user's associated data 42, the reply text 43, and the target object's parameter information 44. The associated data 42 includes but is not limited to the user's associated identification information and / or text content. The target object's parameter information 44 includes the target object's associated price parameters and inventory parameters.

[0116] The data processing method provided in this embodiment generates target reply content by splicing related data, reply text and parameter information of the target object, thereby being able to increase the user's attention to the target reply content on the basis of replying to the text content posted by the user, and enabling the user to further understand the target object.

[0117] Figure 5 A flowchart of a data processing method provided by another embodiment of the present disclosure is provided. Based on any of the above embodiments, Figure 5 As shown, step 204 includes:

[0118] Step 501: Determine a time range for inserting the target reply content into the live broadcast content based on current display parameters of the live broadcast room.

[0119] Step 502: determine whether there is a chance to insert a reply within the time range according to the live broadcast script associated with the live broadcast content, and obtain a processing result.

[0120] Step 503: determine a target response method according to the processing result and the target intention.

[0121] Step 504: perform a reply operation based on the target reply content in accordance with the target reply method.

[0122] In this embodiment, after the target reply content is generated, in order to be able to perform targeted reply operations on the text content posted by each user without affecting the live broadcast process, the reply insertion timing and the target reply method can be determined.

[0123] Optionally, the current display parameters of the live broadcast room can be determined, wherein the display parameters include the number of viewers in the live broadcast room. Different reply insertion time ranges can be set for different display parameters. For example, when the live broadcast room is very popular, in order to avoid interrupting the current live broadcast progress, high-value text content can be determined in the text content. For example, text content with a target intent of non-conversational intent can be determined as high-value text content. Reply to high-value text content at a fast pace. For example, the target reply content can be inserted within 10 seconds after the text content is obtained. When the number of viewers in the live broadcast room is small, reply operations can be performed on most of the text content, and questions, answers and guidance can be conducted in a point-to-point manner. For example, the target reply content can be inserted within 30 seconds after the text is obtained.

[0124] Furthermore, the live broadcast script generally includes the live broadcast process of the live broadcast, so it is possible to determine whether there is an opportunity to insert a reply within the time range based on the live broadcast script associated with the live broadcast content to obtain the processing result.

[0125] In order to make the reply operation for the text content more suitable for the current live broadcast situation and reply to the text content more timely, the current target reply method can be determined based on the target intention associated with the text content and whether there is a reply insertion opportunity. After determining the reply insertion opportunity and the target reply method, the reply operation can be performed based on the target reply content according to the target reply method.

[0126] The data processing method provided in this embodiment determines the timing and method of reply based on the target intention and the current display parameters in the live broadcast room, so that targeted reply operations can be performed on the text content posted by each user without affecting the live broadcast process, thereby improving the user experience.

[0127] Further, based on any of the above embodiments, step 501 includes:

[0128] If the number of viewers in the live broadcast room reaches a preset threshold, the time range is determined as a preset first time range.

[0129] If the number of viewers in the live broadcast room is lower than the threshold value, the time range is determined as a preset second time range.

[0130] The first time range is smaller than the second time range.

[0131] In this embodiment, different reply insertion time ranges may be determined for different display parameters associated with the live broadcast room, wherein the display parameters may include the number of viewers associated with the live broadcast room.

[0132] Optionally, if the number of viewers in the live broadcast room reaches a preset number threshold, the time range is determined as a preset first time range. If the number of viewers in the live broadcast room is lower than the number threshold, the time range is determined as a preset second time range. The first time range is smaller than the second time range.

[0133] For example, when there are many viewers in the live broadcast room, in order to improve the viewing experience of users, the text content can be replied to more promptly. For example, the first time range can be 10 seconds after the user publishes the text content, and the second time range can be 30 seconds after the user publishes the text content. Alternatively, the time range can be determined by the live broadcast room manager according to actual needs, and the present disclosure does not limit this.

[0134] The data processing method provided in this embodiment sets different response time ranges for live broadcast rooms with different numbers of viewers, thereby preventing responses to text content from affecting the live broadcast process, allowing users of the live broadcast room to watch better quality live broadcast content in the live broadcast room, thereby improving user experience.

[0135] Further, based on any of the above embodiments, step 502 includes:

[0136] Identify annotation information associated with at least part of the text paragraphs in the live broadcast script, where the annotation information is used to indicate that there is no opportunity to insert voice content associated with the text paragraphs during playback.

[0137] For a target text paragraph that does not include annotation information in the live broadcast script, a segmentation operation is performed on the target text paragraph according to punctuation marks and / or semantic recognition results associated with the target text paragraph to obtain a segmentation result.

[0138] Based on the segmentation result and the current playback progress of the live broadcast script, determine whether there is a reply insertion opportunity within the time range.

[0139] In this embodiment, the live broadcast script generally includes the live broadcast process of the live broadcast, such as the opening remarks, explanations for each target object, closing remarks, etc. Therefore, the timing of inserting a reply that does not affect the current live broadcast progress can be determined based on the live broadcast script.

[0140] Optionally, during the live broadcast, when explaining the target object, inserting the target reply content often interrupts the current explanation. Alternatively, when explaining based on a semantically complete paragraph, inserting the target reply content often interrupts the current explanation. Alternatively, for live broadcasts with goods, when the "forcing orders" process is in progress, inserting the target reply content often interrupts the current live broadcast progress.

[0141] As a possible implementation method, the paragraphs in the live script that do not have the insertion opportunity can be marked in advance. Therefore, after obtaining the live script, the marking information associated with at least part of the text paragraphs in the live script can be identified, and the marking information is used to indicate that the voice content associated with the text paragraph does not have the insertion opportunity during the playback process.

[0142] Further, after determining that there is no paragraph with an insertion opportunity, for the target text paragraph that does not include annotation information in the live script, the target text paragraph can be segmented according to the punctuation marks and / or semantic recognition results associated with the target text paragraph to obtain a segmentation result. Based on the segmentation result and the current playback progress of the live script, it is determined whether there is a reply insertion opportunity within the time range.

[0143] For example, after determining the current playback progress of the live broadcast script, the reply insertion time closest to the current playback progress can be determined based on the above segmentation results. Determine the time required to play from the current playback progress to the reply insertion time. For example, the number of words to be played from the current playback progress to the reply insertion time can be determined, and the time required to play the voice after converting the number of words to be played into voice can be determined. If the time is within the time range that matches the live broadcast room display parameters, the target reply content can be converted into a reply voice, and the virtual anchor can be controlled to read the reply voice.

[0144] On the contrary, if the duration exceeds the time range that matches the live broadcast room display parameters, in order to be able to reply to the text content posted by the user in a timely manner while avoiding interrupting the live broadcast content, the reply operation can be made through a reply in the comment area or a one-to-one conversation. The present disclosure does not impose any restrictions on this.

[0145] The data processing method provided in this embodiment can avoid interrupting the current live broadcast content by replying operations to the text content by pre-marking the live broadcast script, thereby improving the content quality of the live broadcast.

[0146] Further, based on any of the above embodiments, step 503 includes:

[0147] If the target intention is not a conversational intention and there is a chance for reply insertion, the target reply content is converted into a reply voice, and the preset virtual anchor is controlled to play the reply voice at the reply insertion chance.

[0148] If the target intent is not a conversational intent and there is no opportunity to insert a reply, the reply text in the target reply content is displayed in the comment area.

[0149] If the target intent is not a conversation intent, and the number of times the user has posted text content exceeds a threshold, in response to a conversation request triggered by the user, a one-to-one conversation is conducted with the user based on the target reply content in a preset conversation page, wherein the conversation request is initiated by the user by triggering a conversation control in a display interface, and the conversation control is displayed after the number of times the user has posted text content exceeds the threshold.

[0150] If the target intent is a conversation intent, the reply text in the target reply content is displayed in the comment area within the preset time range.

[0151] In this embodiment, in order to make the reply operation for the text content more suitable for the current live broadcast situation and to reply to the text content more timely, the current target reply method can be determined based on the target intention associated with the text content and whether there is a chance to insert a reply.

[0152] Optionally, if the target intent is not a conversation intent, and there is a chance to insert a reply, that is, the user wants to learn more about the target object, and inserting the target reply content during the live broadcast will not affect the live broadcast progress. Therefore, the target reply content can be converted into a reply voice through a preset voice generation algorithm, and the preset virtual anchor can be controlled to play the reply voice at the reply insertion opportunity. For example, the target reply content can be converted into a reply voice through TTS.

[0153] Optionally, if the target intent is not a conversation intent and there is no opportunity to insert a reply, that is, the user wants to learn more about the target object, but inserting the target reply content during the live broadcast will affect the live broadcast content. Therefore, in order to be able to reply to the text content posted by the user in a timely manner without affecting the live broadcast, the reply text in the target reply content can be displayed in the comment area.

[0154] Optionally, if the target intent is not a conversation intent, and the number of times the user posts text content exceeds the number threshold, that is, the user is more eager to learn more about the target object. Therefore, after the number of times the user posts text content exceeds the number threshold, a conversation control can be displayed in the user's display interface. After the user triggers the conversation control, a one-to-one conversation can be conducted with the user based on the target reply content in a preset conversation page. In addition, after the target reply content is displayed, a subsequent one-to-one conversation can also be conducted with the user through a preset intelligent customer service.

[0155] Optionally, if the target intent is a conversation intent, the reply text in the target reply content is displayed in the comment area within a preset time range.

[0156] The data processing method provided in this embodiment determines the target reply method according to different target intentions and reply insertion timings, so that the current reply timing and reply method for the text content can be more in line with the user's actual needs, thereby improving the user experience.

[0157] Further, based on any of the above embodiments, before identifying the annotation information associated with at least part of the text paragraphs in the live broadcast script, the method further includes:

[0158] Get the script to be processed.

[0159] The paragraphs in the script to be processed that meet the preset annotation conditions are identified through a preset large language model, wherein the annotation conditions include identifying semantically complete paragraphs, identifying paragraphs associated with a preset live broadcast stage, and identifying explanation paragraphs associated with a target object.

[0160] Add annotation information to the paragraphs that meet the preset annotation conditions.

[0161] or,

[0162] Get the script to be processed.

[0163] In response to a marking operation triggered by a user on a script to be processed, a marking position currently selected by the user is determined.

[0164] Add annotation information at the annotation position.

[0165] In this embodiment, in order to determine the timing of reply insertion based on the live broadcast script, it is first necessary to mark the live broadcast script.

[0166] Optionally, the live broadcast script can be annotated by a large language model. The large language model can be pre-input with guidance description information, so that the large language model can recognize semantically complete paragraphs, paragraphs associated with the preset live broadcast stage, and explanation paragraphs associated with the target object, and can annotate the above paragraphs.

[0167] Therefore, after obtaining the script to be processed, the paragraphs in the script to be processed that meet the preset annotation conditions can be identified through the preset large language model, and annotation information can be added to the paragraphs that meet the preset annotation conditions.

[0168] As an implementable method, in order to make the live broadcast content more in line with the actual needs of users, users can also manually add annotation information to the live broadcast script.

[0169] Therefore, a script to be processed can be obtained, wherein the script to be processed may be an unlabeled live broadcast script, or the script to be processed may be a live broadcast script labeled by a large language model.

[0170] Furthermore, the script to be processed can be displayed in the display interface, and the user can trigger a marking operation in the script to be processed according to actual needs. In response to the marking operation, the marking position currently selected by the user can be determined. The marking position can be the position where the cursor stops, or the marking position can be the position manually selected by the user. In response to the adding operation triggered by the user at the marking position, the marking information can be added at the marking position.

[0171] The data processing method provided in this embodiment uses a preset large language model to perform annotation operations on live broadcast scripts, thereby accurately identifying and annotating paragraphs that meet preset annotation conditions, and avoiding the reply content interrupting the current coherent live broadcast content. In addition, by annotating the live broadcast script based on the annotation operation triggered by the user, it is possible to make the insertion points in the live broadcast process more in line with the personalized needs of users on the basis of avoiding the reply content interrupting the current coherent live broadcast content.

[0172] Figure 6 A schematic diagram of the structure of a data processing device provided in an embodiment of the present disclosure, such as Figure 6As shown, the device includes: an acquisition module 61, a generation module 62, a processing module 63 and a reply module 64. Among them, the acquisition module 61 is used to obtain the text content published by the user in the live broadcast room for any target object associated with the live broadcast content, and identify the target intent associated with the text content through a preset intent classification model. The generation module 62 is used to generate a reply text corresponding to the text content using a network model matching the target intent. The processing module 63 is used to generate a target reply content according to the user's associated data, the reply text and the parameter information of the target object, wherein the associated data includes the user's identification information and the text content. The reply module 64 is used to reply to the text content based on the target reply content within the reply opportunity that meets the preset conditions.

[0173] Further, based on any of the above embodiments, the device further includes: an acquisition module for acquiring a preset training data set, wherein the training data set includes a plurality of training data pairs, the training data pairs include training texts and annotation information associated with the training texts, and the annotation information includes intent information corresponding to the training texts. A training module for iteratively training a preset classification model using the training data set until the classification model meets a preset convergence condition, thereby obtaining the intent classification model.

[0174] Further, based on any of the above embodiments, the network model includes any one of a large language model and at least one small model associated with the large language model. The device also includes: a training module, which is used to input preset text description information into the large language model, the text description information includes parameter information of at least one target object associated with the live content, a live script, a preset question and answer database, and condition information associated with the reply text, and the text description information is used to guide the large language model to generate a reply text that meets the condition information based on the parameter information, the live script, and the preset question and answer database.

[0175] Further, on the basis of any of the above embodiments, the target intent includes parameter query intent, descriptive information consultation intent and conversation intent associated with the target object. The generation module includes: a first generation unit, which is used to input the text content into a preset large language model if the target intent is a conversation intent, and obtain the reply text output by the large prediction model. A second generation unit, which is used to input the text content into a small query model associated with the large language model if the target intent is a parameter query intent, and obtain the reply text output by the small query model. A third generation unit, which is used to input the text content into a small text generation model associated with the large language model if the target intent is a descriptive information consultation intent, and obtain the reply text output by the small text generation model.

[0176] Further, based on any of the above embodiments, the processing module includes: a first determination unit, used to use the identification information associated with the user and / or the text content as the initial text. A second determination unit, used to use the reply text as the intermediate text and the parameter information associated with the target object as the end text, wherein the parameter information includes the price parameter and inventory parameter associated with the target object. A splicing unit, used to sequentially splice the initial text, the intermediate text and the end text to obtain the target reply content.

[0177] Further, on the basis of any of the above embodiments, the reply module includes: a determination unit, configured to determine a time range for inserting the target reply content into the live content based on the current display parameters of the live broadcast room. A judgment unit, configured to determine whether there is a reply insertion opportunity within the time range according to the live broadcast script associated with the live broadcast content, and obtain a processing result. A processing unit, configured to determine a target reply method according to the processing result and the target intention. A reply unit, configured to perform a reply operation based on the target reply content according to the target reply method.

[0178] Further, based on any of the above embodiments, the determining unit includes: a first determining subunit, configured to determine the time range as a preset first time range if the number of viewers in the live broadcast room reaches a preset number threshold. A second determining subunit, configured to determine the time range as a preset second time range if the number of viewers in the live broadcast room is lower than the number threshold. The first time range is smaller than the second time range.

[0179] Further, on the basis of any of the above embodiments, the judgment unit includes: an identification subunit, used to identify annotation information associated with at least part of the text paragraphs in the live broadcast script, wherein the annotation information is used to characterize that there is no opportunity to insert the voice content associated with the text paragraph during the playback process. A segmentation subunit, used to segment the target text paragraph that does not include annotation information in the live broadcast script according to the punctuation marks and / or semantic recognition results associated with the target text paragraph to obtain a segmentation result. A judgment subunit, used to determine whether there is an opportunity to insert a reply within the time range based on the segmentation result and the current playback progress of the live broadcast script.

[0180] Further, on the basis of any of the above embodiments, the processing unit includes: a first processing subunit, which is used to convert the target reply content into a reply voice if the target intention is not a conversation intention and there is a reply insertion opportunity, and control the preset virtual anchor to play the reply voice at the reply insertion opportunity. A second processing subunit, which is used to display the reply text in the target reply content in the comment area if the target intention is not a conversation intention and there is no reply insertion opportunity. A third processing subunit, which is used to respond to a conversation request triggered by the user, and conduct a one-to-one conversation with the user based on the target reply content in a preset conversation page if the target intention is not a conversation intention and the number of times the user posts text content exceeds the number threshold, wherein the conversation request is initiated by the user by triggering the conversation control in the display interface, and the conversation control is displayed after the number of times the user posts text content exceeds the number threshold. A fourth processing subunit, which is used to display the reply text in the target reply content in the comment area within a preset time range if the target intention is a conversation intention.

[0181] Further, on the basis of any of the above embodiments, the device also includes: an acquisition module for acquiring the script to be processed. An identification module for identifying paragraphs in the script to be processed that meet preset annotation conditions through a preset large language model, wherein the annotation conditions include identifying semantically complete paragraphs, identifying paragraphs associated with a preset live broadcast stage, and identifying explanation paragraphs associated with a target object. An adding module for adding annotation information to paragraphs that meet preset annotation conditions. Alternatively, an acquisition module for acquiring the script to be processed. A determination module for determining the annotation position currently selected by the user in response to an annotation operation triggered by the user for the script to be processed. An adding module for adding annotation information at the annotation position.

[0182] Further, based on any of the above embodiments, the device further comprises: a storage module for associating and storing the text content and the target reply content generated based on the text content to obtain an optimized data pair; and an optimization module for optimizing the network model based on the optimized data pair within a preset time range.

[0183] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0184] According to an embodiment of the present disclosure, the present disclosure further provides an electronic device, including:

[0185] at least one processor; and

[0186] a memory communicatively connected to the at least one processor; wherein,

[0187] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described in any one of the above embodiments.

[0188] According to an embodiment of the present disclosure, the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the method described in any one of the above embodiments.

[0189] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0190] Figure 7 A schematic diagram of the structure of an electronic device provided for an embodiment of the present disclosure. The electronic device 700 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0191] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0192] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0193] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as data processing methods. For example, in some embodiments, the data processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the data processing method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the data processing method in any other appropriate manner (e.g., by means of firmware).

[0194] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0195] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0196] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0197] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0198] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0199] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0200] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0201] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A data processing method, comprising: Acquire text content posted by a user in the live broadcast room for any target object associated with the live broadcast content, and identify the target intent associated with the text content through a preset intent classification model; Using a network model that matches the target intent to generate a reply text corresponding to the text content; Generate target reply content according to the user's associated data, the reply text and the parameter information of the target object, wherein the associated data includes the user's identification information and the text content; A reply operation is performed on the text content based on the target reply content within a reply opportunity that meets preset conditions.

2. The method according to claim 1, before identifying the target intent associated with the text content by a preset intent classification model, further comprising: Acquire a preset training data set, wherein the training data set includes a plurality of training data pairs, the training data pairs include training texts and annotation information associated with the training texts, and the annotation information includes intent information corresponding to the training texts; The preset classification model is iteratively trained using the training data set until the classification model meets the preset convergence condition, thereby obtaining the intent classification model.

3. The method according to claim 1, wherein the network model comprises any one of a large language model and at least one small model associated with the large language model; before the network model matching the target intent is used to generate a reply text corresponding to the text content, the method further comprises: Preset text description information is input into the large language model, the text description information including parameter information of at least one target object associated with the live content, a live script, a preset question and answer database, and condition information associated with the reply text, and the text description information is used to guide the large language model to generate a reply text that meets the condition information based on the parameter information, the live script, and the preset question and answer database.

4. The method according to claim 3, wherein: The target intent includes parameter query intent, description information consultation intent and conversation intent associated with the target object; The step of using a network model that matches the target intent to generate a reply text corresponding to the text content includes: If the target intent is a conversation intent, the text content is input into a preset large language model to obtain a reply text output by the large prediction model; If the target intent is a parameter query intent, the text content is input into the small query model associated with the large language model to obtain the reply text output by the small query model; If the target intent is to describe information consultation intent, the text content is input into the small text generation model associated with the large language model to obtain the reply text output by the small text generation model.

5. The method according to claim 1, wherein: The generating target reply content according to the associated data of the user, the reply text and the parameter information of the target object includes: Using the identification information associated with the user and / or the text content as the initial text; The reply text is used as the intermediate text, and the parameter information associated with the target object is used as the end text, wherein the parameter information includes the price parameter and the inventory parameter associated with the target object; The initial text, the intermediate text and the end text are concatenated in sequence to obtain the target reply content.

6. The method according to claim 1, wherein: The step of replying to the text content based on the target reply content within a reply opportunity that meets a preset condition includes: Determine a time range for inserting the target reply content into the live content based on current display parameters of the live broadcast room; Determine whether there is a reply insertion opportunity within the time range according to the live broadcast script associated with the live broadcast content, and obtain a processing result; Determine a target response method according to the processing result and the target intention; Perform a reply operation according to the target reply method and based on the target reply content.

7. The method according to claim 6, wherein: The determining, based on the current display parameters of the live broadcast room, a time range for inserting the target reply content into the live broadcast content includes: If the number of viewers in the live broadcast room reaches a preset threshold, the time range is determined as a preset first time range; If the number of viewers in the live broadcast room is lower than the number threshold, the time range is determined as a preset second time range; The first time range is smaller than the second time range.

8. The method according to claim 6, wherein: Determining whether there is a reply insertion opportunity within the time range according to the live broadcast script associated with the live broadcast content, and obtaining a processing result, including: Identifying annotation information associated with at least part of the text paragraphs in the live broadcast script, wherein the annotation information is used to indicate that there is no opportunity to insert voice content associated with the text paragraphs during playback; For a target text paragraph that does not include annotation information in the live broadcast script, segment the target text paragraph according to punctuation marks and / or semantic recognition results associated with the target text paragraph to obtain a segmentation result; Based on the segmentation result and the current playback progress of the live broadcast script, determine whether there is a reply insertion opportunity within the time range.

9. The method according to claim 6, wherein: The determining of the target response method according to the processing result and the target intention includes: If the target intention is not a conversation intention and there is a reply insertion opportunity, the target reply content is converted into a reply voice, and the preset virtual anchor is controlled to play the reply voice at the reply insertion opportunity; If the target intent is not a conversation intent and there is no opportunity to insert a reply, the reply text in the target reply content is displayed in the comment area; If the target intent is not a conversation intent, and the number of times the user has posted text content exceeds a threshold, in response to a conversation request triggered by the user, a one-to-one conversation is conducted with the user based on the target reply content in a preset conversation page, wherein the conversation request is initiated by the user by triggering a conversation control in the display interface, and the conversation control is displayed after the number of times the user has posted text content exceeds the threshold; If the target intent is a conversation intent, the reply text in the target reply content is displayed in the comment area within the preset time range.

10. The method according to claim 6, before identifying the annotation information associated with at least part of the text paragraphs in the live broadcast script, further comprising: Get the script to be processed; Identifying paragraphs in the script to be processed that meet preset annotation conditions through a preset large language model, wherein the annotation conditions include identifying semantically complete paragraphs, identifying paragraphs associated with a preset live broadcast stage, and identifying explanation paragraphs associated with a target object; Add annotation information to the paragraphs that meet the preset annotation conditions; or, Get the script to be processed; In response to a marking operation triggered by a user on a script to be processed, determining a marking position currently selected by the user; Add annotation information at the annotation position.

11. The method according to any one of claims 1 to 10, further comprising: Associatively storing the text content and the target reply content generated based on the text content to obtain an optimized data pair; The network model is optimized based on the optimization data within a preset time range.

12. A data processing device, comprising: An acquisition module is used to acquire text content published by a user in a live broadcast room for any target object associated with the live broadcast content, and identify the target intent associated with the text content through a preset intent classification model; A generation module, used to generate a reply text corresponding to the text content by using a network model matching the target intent; A processing module, configured to generate target reply content according to the associated data of the user, the reply text and parameter information of the target object, wherein the associated data includes the identification information of the user and the text content; The reply module is used to reply to the text content based on the target reply content within a reply opportunity that meets preset conditions.

13. The device according to claim 12, further comprising: An acquisition module, used to acquire a preset training data set, wherein the training data set includes a plurality of training data pairs, the training data pairs include training texts and annotation information associated with the training texts, and the annotation information includes intent information corresponding to the training texts; The training module is used to iteratively train the preset classification model through the training data set until the classification model meets the preset convergence condition, thereby obtaining the intent classification model.

14. The apparatus according to claim 12, wherein the network model comprises any one of a large language model and at least one small model associated with the large language model; the apparatus further comprises: A training module is used to input preset text description information into the large language model, wherein the text description information includes parameter information of at least one target object associated with the live content, a live script, a preset question and answer database, and condition information associated with the reply text, and the text description information is used to guide the large language model to generate a reply text that meets the condition information based on the parameter information, the live script, and the preset question and answer database.

15. The device according to claim 14, wherein: The target intent includes parameter query intent, description information consultation intent and conversation intent associated with the target object; The generation module comprises: A first generating unit is configured to input the text content into a preset large language model to obtain a reply text output by the large prediction model if the target intent is a conversation intent; A second generating unit is configured to input the text content into a small query model associated with the large language model if the target intent is a parameter query intent, and obtain a reply text output by the small query model; The third generation unit is used to input the text content into the text generation small model associated with the large language model if the target intention is to describe information consultation intention, and obtain the reply text output by the text generation small model.

16. The device according to claim 12, wherein: The processing module comprises: A first determining unit, configured to use the identification information associated with the user and / or the text content as the initial text; A second determining unit is used to use the reply text as an intermediate text and parameter information associated with the target object as an end text, wherein the parameter information includes a price parameter and an inventory parameter associated with the target object; A concatenation unit is used to sequentially concatenate the initial text, the intermediate text and the end text to obtain the target reply content.

17. The device according to claim 12, wherein: The reply module comprises: A determination unit, configured to determine a time range for inserting the target reply content into the live broadcast content based on current display parameters of the live broadcast room; A judgment unit, configured to determine whether there is a reply insertion opportunity within the time range according to a live broadcast script associated with the live broadcast content, and obtain a processing result; A processing unit, configured to determine a target response method according to the processing result and the target intention; A reply unit is used to perform a reply operation based on the target reply content in accordance with the target reply method.

18. The device according to claim 17, wherein: The determining unit comprises: A first determining subunit is configured to determine the time range as a preset first time range if the number of viewers in the live broadcast room reaches a preset number threshold; A second determining subunit is configured to determine the time range as a preset second time range if the number of viewers in the live broadcast room is lower than the number threshold; The first time range is smaller than the second time range.

19. The device according to claim 17, wherein: The judging unit comprises: An identification subunit, used to identify annotation information associated with at least part of the text paragraphs in the live broadcast script, wherein the annotation information is used to indicate that there is no insertion opportunity for the voice content associated with the text paragraphs during the playback process; A segmentation subunit is used to segment the target text paragraph that does not include annotation information in the live broadcast script according to punctuation marks and / or semantic recognition results associated with the target text paragraph to obtain a segmentation result; A judgment subunit is used to determine whether there is a reply insertion opportunity within the time range based on the segmentation result and the current playback progress of the live broadcast script.

20. The device according to claim 17, wherein: The processing unit comprises: The first processing sub-unit is configured to convert the target reply content into a reply voice if the target intention is not a conversation intention and there is a reply insertion opportunity, and control a preset virtual anchor to play the reply voice at the reply insertion opportunity; The second processing subunit is configured to display the reply text in the target reply content in the comment area if the target intention is not a conversation intention and there is no reply insertion opportunity; a third processing subunit, configured to, if the target intent is not a conversation intent and the number of times the user has posted text content exceeds a number threshold, respond to a conversation request triggered by the user, conduct a one-to-one conversation with the user based on the target reply content in a preset conversation page, wherein the conversation request is initiated by the user by triggering a conversation control in the display interface, and the conversation control is displayed after the number of times the user has posted text content exceeds the number threshold; The fourth processing sub-unit is used to display the reply text in the target reply content in the comment area within a preset time range if the target intention is a conversation intention.

21. The apparatus according to claim 17, further comprising: The acquisition module is used to obtain the script to be processed; An identification module, used to identify paragraphs in the script to be processed that meet preset annotation conditions through a preset large language model, wherein the annotation conditions include identifying semantically complete paragraphs, identifying paragraphs associated with a preset live broadcast stage, and identifying explanation paragraphs associated with a target object; An adding module is used to add annotation information to paragraphs that meet preset annotation conditions; or, The acquisition module is used to obtain the script to be processed; A determination module, configured to determine a currently selected annotation position of the user in response to an annotation operation triggered by the user on the script to be processed; The adding module is used to add annotation information at the annotation position.

22. The device according to any one of claims 12 to 21, further comprising: A storage module, used for associating and storing the text content and the target reply content generated based on the text content to obtain an optimized data pair; The optimization module is used to optimize the network model based on the optimization data within a preset time range.

23. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 11.

24. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-11.

25. A computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 11.

Images