Data processing and abstract generation method, medium, computer equipment and program product

Abstract

The invention discloses a data processing and abstract generation method, a medium, computer equipment and a program product. The method comprises the following steps: acquiring user data of potential users of a service platform acquired in a historical time period; taking the user data as input data of a large language model, and taking at least one pre-generated feature tag as prompt information of the large language model, so as to extract feature data corresponding to the at least one feature tag from the user data through the large language model; the at least one feature tag is related to the admission willingness of the potential user to a target service provided by the service platform; and generating description information corresponding to the potential user according to the at least one feature tag and the feature data corresponding to the at least one feature tag, wherein the description information is used for describing the admission willingness of the potential user to the target service.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to data processing and summary generation methods, media, computer devices and program products. Background Art

[0002] Mining and following up potential users is a crucial part of sales work. On the service platform, members of the customer service team often need to process a large amount of user data of potential users and ensure that each potential user can receive personalized and effective follow-up. However, due to the large amount of user data, members of the customer service team need to spend a lot of time to understand and process user data, resulting in low follow-up efficiency. Summary of the invention

[0003] In a first aspect, an embodiment of the present application provides a data processing method, the method comprising: obtaining user data of potential users of a service platform collected within a historical time period; using the user data as input data of a large language model, and using at least one pre-generated feature tag as prompt information of the large language model, so as to extract feature data corresponding to the at least one feature tag from the user data through the large language model; the at least one feature tag is related to the potential user's willingness to accept a target service provided by the service platform; and generating description information corresponding to the potential user based on the at least one feature tag and its corresponding feature data, the description information being used to describe the potential user's willingness to accept the target service.

[0004] In a second aspect, an embodiment of the present application provides a summary generation method, the method comprising: obtaining user data of potential users of an e-commerce platform collected within a historical time period; using the user data as input data of a large language model, and using at least one pre-generated feature tag as prompt information of the large language model, so as to extract feature data corresponding to the at least one feature tag from the user data through the large language model; the at least one feature tag is related to the potential user's willingness to accept the e-commerce service provided by the e-commerce platform; generating a summary corresponding to the potential user based on the at least one feature tag and its corresponding feature data, the summary being used to describe the potential user's willingness to accept the e-commerce service.

[0005] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present application.

[0006] In a fourth aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present application when executing the program.

[0007] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the method described in any embodiment of the present application.

[0008] In an embodiment of the present application, after obtaining the user data of the potential user, the feature data that affects the potential user's willingness to accept the target service provided by the service platform is extracted from the user data through a large language model, and descriptive information is further generated based on the feature data and its feature tags to describe the potential user's willingness to accept the target service. When following up with potential users, members of the customer service team do not need to spend a lot of time and energy to process massive amounts of user data, but can directly and intuitively obtain descriptive information describing the potential user's willingness to accept based on the output of the large language model, thereby improving the efficiency of the customer service team members in following up with potential users.

[0009] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings herein are incorporated into the specification and constitute a part of the present application. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solutions of the present application.

[0011] Figure 1 It is a schematic diagram of an application scenario of an embodiment of the present application.

[0012] Figure 2 It is a flow chart of the data processing method of an embodiment of the present application.

[0013] Figure 3 It is a schematic diagram of the source of user data in an embodiment of the present application.

[0014] Figure 4 It is a schematic diagram of the heterogeneous data integration process of an embodiment of the present application.

[0015] Figure 5 It is a schematic diagram of the process of automatically generating prompt words in an embodiment of the present application.

[0016] Figure 6 It is a schematic diagram of the process of determining the changing trend of the willingness to accept according to an embodiment of the present application.

[0017] Figure 7 It is a system architecture diagram of an embodiment of the present application.

[0018] Figure 8 It is an overall flow chart of an embodiment of the present application.

[0019] Fig. 9 It is a flowchart of the summary generation method of an embodiment of the present application.

[0020] Fig.10 It is a schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0022] The terms used in this application are only for the purpose of describing specific embodiments, and are not intended to limit the application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a variety of.

[0023] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0024] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application and to make the above-mentioned purposes, features and advantages of the embodiments of the present application more obvious and understandable, the technical solutions in the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0025] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0026] Mining and following up potential users is a crucial part of sales work. On the service platform, members of the customer service team often need to process a large amount of user data of potential users and ensure that each potential user can receive personalized and effective follow-up. Figure 1 An example of a possible application scenario of the embodiment of the present application is given. It should be understood that the application scenarios listed below are only exemplary and are not intended to limit the present application. Figure 1 In addition to the application scenarios shown, the present application can also be applied to other application scenarios. Figure 1 As shown, the service platform 10 can provide target services to the potential users 30 of the service platform 10, and the salesperson 204 of the service platform 10 can communicate with the potential users 30 of the service platform 10 through offline interviews and other methods, and record the user data of the potential users 30 during the communication process. These user data can be summarized to the service platform 10. Since the time of following up the potential users 30 may be relatively long, the user data generated will also be relatively large. The salesperson 204 may need to continuously summarize the user data in order to understand the changing trend of the potential user 30's willingness to accept the target service provided by the service platform 10. At the same time, after the salesperson 204 resigns or transfers, the new salesperson 204 needs to continue to follow up the potential user 30. The new salesperson 204 needs to organize the user data of the potential user 30 that has been currently acquired in order to quickly understand the situation of the potential user 30. In the above process, the salesperson needs to spend a lot of time to understand and process the user data, resulting in low follow-up efficiency.

[0027] Based on this, the embodiment of the present application provides a data processing method, which uses a large language model to summarize user data to obtain description information corresponding to potential users, which is used to describe the potential user 30's willingness to accept the target service. Figure 1 In the embodiment shown, when following up with potential users 30, salespersons 202 and 204 do not need to spend a lot of time and energy to process massive amounts of user data, but can directly obtain description information describing the acceptance willingness of potential users 30 based on the output of the large language model, thereby improving the efficiency of the customer service team members in following up with potential users. Figure 1 Taking the application scenario shown as an example, the solution of the embodiment of the present application is illustrated.

[0028] like Figure 2 As shown, the data processing method of the embodiment of the present application includes:

[0029] Step S12: obtaining user data of potential users of the service platform 10 collected within a historical time period;

[0030] Step S14: using the user data as input data of the large language model, and using at least one pre-generated feature tag as prompt information of the large language model, so as to extract feature data corresponding to at least one feature tag from the user data through the large language model; the at least one feature tag is related to the willingness of the potential user 30 to accept the target service provided by the service platform 10;

[0031] Step S16: Generate description information corresponding to the potential user 30 according to at least one feature tag and its corresponding feature data, where the description information is used to describe the potential user 30's willingness to accept the target service.

[0032] In step S12, the service platform 10 refers to a platform that can provide the target service. For example, when the target service is an e-commerce service, the service platform 10 is an e-commerce platform; when the target service is an e-book reading service, the service platform 10 is an e-book reading platform; when the target service is a video playback service, the service platform 10 is a video playback platform; when the target service is a news reading service, the service platform 10 is a news reading platform. The target service and the service platform 10 can also be other types, which are not listed here one by one. For the sake of ease of description, the following is an example in which the target service is an e-commerce service and the service platform 10 is an e-commerce platform.

[0033] Potential users refer to a group of users who may become users of the service platform 10, that is, users who may be interested in the target services provided by the service platform 10. Potential users can be determined manually. For example, members of the customer service team can obtain the user's contact information through various channels, and determine the users who can obtain the contact information as potential users. Alternatively, further communication can be conducted with the user whose contact information can be obtained. If it is determined based on the communication results that the user is interested in the target service provided by the service platform 10, the user is determined to be a potential user. Potential users can also be determined based on the user's browsing history on websites related to the service platform 10 or the user's behavior sequence on the service platform 10. Potential users can also be determined automatically by software, which is not limited by this application.

[0034] The user data collected in the historical time period may be obtained, and the user data is the data of potential users of the service platform 10. The potential user 30 may be an individual user or an enterprise user.

[0035] The historical time period may include all time before the current moment, or may only include time whose time interval with the current time is within a preset time interval range. For example, the preset time interval may be a time interval of 1 month, 3 months, half a year, or 1 year. Accordingly, the historical time period may include time whose time interval with the current time is within 1 month, 3 months, half a year, or 1 year. Since user data may change, only using user data whose time interval with the current time is within the preset time interval range can reduce data inaccuracy caused by changes in user data, thereby improving the accuracy of processing results.

[0036] like Figure 3 As shown, the above data collection process may include an automatically executed data collection process, or may include a manually executed data collection process.

[0037] In an embodiment of automatically collecting user data, for example, when a potential user 30 accesses the service platform 10, user data will be generated, and the user data can be automatically recorded in the database of the service platform 10. For example, in addition to providing the target service, the service platform 10 can also provide other services (such as logistics services). Potential users 30 can generate user data in the process of using other services provided by the service platform 10, and the user data can be recorded in the database. The user data of the potential user 30 can be queried from the database through a structured query statement. This process is automated, which ensures the seamless integration of information between other services generated by the service platform 10 and the target service, and provides a real-time and accurate data stream for the analysis of user willingness. The service platform 10 can pre-establish an interface with the database to obtain user data from the database through the interface.

[0038] For another example, when a potential user 30 accesses the Internet (e.g., official websites, recruitment platforms, business information websites, and industry-related professional websites, etc.) through his / her own identification information (e.g., ID card number, mobile phone number, or account number, etc.), the user data of the potential user 30 can be collected through a web page tracker on the Internet. In this way, the business activities, expansion plans, industry dynamics, and human resource changes of the potential user 30 can be discovered. After obtaining these user data, the system will analyze the user data to reveal market opportunities and potential business partnerships, so that the customer service team can more accurately locate target customers and optimize sales strategies.

[0039] For another example, a potential user 30 may perform certain actions on his application, and the behavior of the potential user 30 may be simulated through the Robotic Process Automation (RPA) process to obtain user data. For example, a user may perform operations such as adding friends, sending files, browsing user sharing and publishing content on his application. RPA technology can simulate the above operations, thereby efficiently identifying and recording the user's activity level and content preferences.

[0040] For another example, a potential user 30 may visit a website related to the target service and perform certain operations on the website. The data generated when the potential user 30 performs operations on the above website can be obtained through a hook on the website. These hooks are activated when specific operations (such as collection, like, forwarding, purchase, etc.) occur on the internal website. This means that any sales activities or user feedback can be automatically recorded and included in the analysis scope, ensuring that the latest user data is obtained in the decision-making process. The service platform 10 can pre-establish an interface between websites related to the target service and obtain user data from the website through the interface.

[0041] In an embodiment of the data collection process performed manually, user data can be manually uploaded to the service platform 10 by a member of the customer service team. Specifically, members of the customer service team can communicate with potential users 30 through offline visits, telephone calls, emails, or instant messaging tools, and user data can be generated during the communication process. Members of the customer service team can manually upload the user data generated in the above process to the service platform 10. In addition, since the automatically collected data may have certain errors and losses, the automatically collected data can also be corrected and supplemented manually. Manually input data plays a supplementary role. Through manually recorded data, details that may be missed by the automated system can be captured, such as personalized customer communications and specific sales opportunities. These data help to build a more comprehensive user portrait and enhance sales strategies. For example, the service platform 10 can provide a user interface through which users can upload user data.

[0042] In the above embodiment, obtaining user data from multiple data sources can improve the comprehensiveness of the obtained user data, so that the subsequent process can more comprehensively analyze the potential user 30's willingness to accept the target service, thereby improving the accuracy of the analysis.

[0043] Since the user data of the potential user 30 may come from multiple data sources, in some embodiments, the time information corresponding to the user data collected from the multiple data sources may be obtained, and the user data collected from the multiple data sources may be time-series aligned based on the time information. The above time-series alignment refers to sorting and integrating the user data collected from multiple data sources to form user data with a unified timeline. Figure 4 As shown, assuming that there are three data sources, respectively denoted as data source 1, data source 2, and data source 3, and the user data of each data source has a timestamp, the user data of each data source can be time-series aligned according to the timestamp of the user data of each data source. Figure 4 It can be seen that the aligned user data merges the user data from multiple data sources and sorts the merged data according to the timestamp of the user data.

[0044] In step S14, the user data may be input into a large language model (LLM) as input data of the large language model. A large language model is a language model trained using a large-scale corpus, which has a powerful ability to understand and generate natural language. In addition, at least one feature tag may be pre-generated as prompt information for the large language model to guide the large language model to extract feature data corresponding to at least one feature tag from the user data.

[0045] The at least one feature tag mentioned above is related to the potential user's 30 willingness to accept the target service provided by the service platform 10 , that is, the at least one feature tag mentioned above can affect the potential user's 30 willingness to accept the target service provided by the service platform 10 .

[0046] When the potential user 30 is an individual user, the at least one feature tag may include but is not limited to some or all of the following: user name, phone number, email address, occupation, age, work status (such as retired, employed, resigned, on vacation, etc.), home address, income, experience in using the service platform 10 (whether the service provided by the service platform 10 has been used, including the target service and / or other services other than the target service), user behavior tags, etc. When the potential user 30 is an enterprise user, the at least one feature tag may include but is not limited to some or all of the following: name, scale, business scope, name of the person in charge, phone number of the person in charge, enterprise address, experience in using the service platform 10, user behavior tags, etc. Among them, the user behavior tags include but are not limited to the user's browsing history of web pages, operation history, activities participated by the user, and / or exhibitions participated by the user.

[0047] The at least one feature label can be obtained manually, but since it is time-consuming and laborious to obtain the feature label manually, and the effect of the obtained feature label may not be good enough, the at least one feature label can also be obtained through a pre-trained label generation model. The label generation model can also be a large language model.

[0048] The labels initially generated by the label generation model may have poor effects, so the labels initially generated by the label generation model may be fine-tuned to obtain at least one feature label input to the large language model in step S14. The following is an example of how to fine-tune the labels.

[0049] In some embodiments, see Figure 5 , the sample data and the sample label corresponding to the sample data can be obtained, and the sample label is used to indicate whether the user to whom the sample data belongs is willing to accept the target service. Then, the current prompt word output by the label generation model based on the sample data and its corresponding sample label can be obtained, and the current prompt word includes at least one current predicted label, and the current performance index corresponding to the current prompt word is determined, and the current performance index is used to evaluate the accuracy of at least one current predicted label included in the current prompt word.

[0050] In this embodiment, the sample label corresponding to the sample data can be obtained by manual labeling. The sample label is used to indicate whether the user to whom the sample data belongs is willing to accept the target service, that is, to indicate whether the sample data is a positive sample or a negative sample. If the user to whom the sample data belongs is willing to accept the target service, the sample data is a positive sample, otherwise the sample data is a negative sample. The label generation model can predict the factors that affect the user's willingness to accept the target service based on the sample data and its corresponding sample label, thereby outputting the current prompt word. The current performance indicator corresponding to the current prompt word can be determined, for example, the recall rate. The recall rate can measure the proportion of topics or key information contained in the prompt word generated by the label generation model. A higher recall rate usually means that the prompt word generated by the model is more likely to cover the content of interest to the user, thereby improving the quality and effect of the feature data subsequently output by the large language model in step S14.

[0051] In some embodiments, the prompt word needs to be verified. The specific verification method is as follows: if the difference between the current performance index and the historical performance index (i.e., the performance index corresponding to the last prompt word output by the label generation model) is greater than the preset threshold, it means that the performance of the prompt word obtained this time is significantly improved compared to the last prompt word obtained, and further attempts can be made to adjust the prompt word to further improve the performance of the prompt word. Therefore, if the difference between the current performance index and the historical performance index is greater than the preset threshold, the step of obtaining the current prompt word output by the label generation model based on the sample data can be returned.

[0052] If the difference between the current performance index and the historical performance index is not greater than the preset threshold, it means that even if the prompt word is adjusted continuously, the performance will not be improved much. Therefore, at least one current predicted label can be determined as at least one feature label input to the large language model in step S14, as the prompt information of the large language model.

[0053] Furthermore, if the difference between the current performance index and the historical performance index is not greater than a preset threshold, the label stability of the sample data can also be obtained, and the label stability is used to evaluate the consistency of the predicted labels output by the label generation model for the sample data multiple times. Specifically, the predicted labels output by the label generation model for the sample data multiple times in a row can be obtained. If the predicted labels output by the label generation model for the same sample data multiple times in a row are inconsistent, it means that the label stability of the sample data is poor, otherwise it means that the label stability of the sample data is good. The more frequently the predicted labels output by the label generation model for the same sample data multiple times in a row change, the worse the label stability of the sample data.

[0054] In some embodiments, if the label stability is higher than a preset stability threshold, at least one current predicted label can be determined as at least one feature label input to the large language model in step S14. If the label stability is not higher than the preset stability threshold, it means that the sample label may be noisy and has low accuracy. Therefore, the sample label can be adjusted (the above process is called noise label cleaning). After adjusting the sample label, it is also possible to return to the step of obtaining the current prompt word output by the label generation model based on the sample data and its corresponding sample label, until the difference between the current performance index and the historical performance index is no greater than the preset threshold, and the label stability of the sample data is higher than the preset stability threshold.

[0055] Understandably, Figure 5 In the illustrated embodiment, different models (such as model A, model B and model C as shown in the figure) are used for prompt word generation, prompt word verification and noise label cleaning respectively. However, in actual applications, the above functions can also be integrated into the same model, that is, model A, model B and model C can be the same model.

[0056] After determining at least one feature tag in the above manner, the feature data corresponding to the at least one feature tag can be extracted through the large language model. The feature tag can be regarded as the key in the key-value pair, and the feature data can be regarded as the value in the key-value pair. For example, when the feature tag includes the name of the enterprise user, the feature data can be "XX Co., Ltd."; when the feature tag includes the name of the person in charge, the feature data can be "Zhang San"; when the feature tag includes the phone number of the person in charge, the feature data can be "18812345678".

[0057] After extracting the feature data corresponding to at least one feature tag, the at least one feature tag and its corresponding feature data may be associated and organized into a key-value pair.

[0058] In step S16, description information corresponding to the potential user can be generated according to at least one feature tag and its corresponding feature data. The description information is used to describe the potential user's willingness to accept the target service. In some embodiments, the large language model can convert at least one feature tag and its corresponding feature data into natural language to obtain description information that conforms to the natural language expression habits, which is convenient for members of the customer service team to read and understand. Among them, the description information may include multiple semantic fragments (for example, a sentence can be used as a semantic fragment). If the number of feature tags is greater than 1, during the conversion, multiple feature tags and their corresponding feature data can be integrated into one semantic fragment to simplify the expression and improve the reading efficiency of members of the customer service team. For example, the feature tag includes the name and the number of visits to the specified web page in the past week, and the corresponding feature data are "Zhang San" and "3" respectively. Then, the above feature tag and its corresponding feature data can be converted into the following description information: "Zhang San visited the specified web page 3 times in the past week".

[0059] In some embodiments, the number of user data is greater than 1. Multiple pieces of user data may be grouped. For example, multiple pieces of user data may be grouped according to the time when the user data was generated. Specifically, each piece of user data may correspond to a timestamp, and multiple pieces of user data may be grouped according to the timestamp corresponding to the user data. Among them, any group includes at least one piece of user data, and the at least one piece of user data may be user data from the same or different data sources, and each piece of user data in the same group is user data generated within the same time period. The time period may be 1 day, 1 week, 1 month, etc. Taking the time period of 1 week as an example, the user data generated within the last week may be divided into one group, and the user data generated between the last 2 weeks and the last week may be divided into another group, and so on. For ease of processing, the multiple pieces of user data may also be sorted according to the timestamps before grouping.

[0060] Alternatively, when the user data includes operations performed by the user, multiple pieces of user data may be grouped according to the operation type. For example, if the operations performed by the user include accessing a web page, registering, signing in, etc., the user data including the operation of accessing a web page may be grouped into one group, the user data including the operation of registering may be grouped into another group, and the user data including the operation of signing in may be grouped into another group.

[0061] User data may also be grouped according to other conditions (eg, business scope of enterprise users, occupation or age of individual users, etc.), which are not listed here one by one.

[0062] In the example of grouping user data according to time periods, at least one group of user data can be obtained, wherein any one group includes at least one piece of user data, and each piece of user data in the same group is user data generated in the same time period. For any one of the at least one group, the description information of the potential user 30 in the time period corresponding to the group can be determined based on at least one feature tag and feature data corresponding to the at least one feature tag extracted from the user data of the group. Then, based on the description information of the potential user 30 in the time period corresponding to the at least one group, the change trend of the potential user 30's willingness to accept the target service is determined.

[0063] Furthermore, if at least one feature tag includes a user behavior tag, which is used to describe the operation performed by the potential user 30 on an object related to the target service, for any one of the at least one groups, the number of feature data corresponding to the specified user behavior tag extracted from the user data of the group can be obtained, and based on the number of feature data corresponding to the specified user behavior tag extracted from the user data of each group, the change trend of the potential user 30's willingness to accept the target service is determined. The operation described by the specified user behavior tag can be an access operation, a browsing operation, or other operation on a specific web page related to the target service. Accordingly, the feature data corresponding to the specified user behavior tag can be feature data corresponding to the access operation and feature data corresponding to the browsing operation.

[0064] Combine the following Figure 6 Taking the operation described by the specified user behavior tag as the access operation to a specific web page as an example, the process of determining the changing trend of the willingness to accept is explained. For the sake of description, it is assumed that at least one feature tag includes the number of visits to a specific web page related to the target service, the number of groups is 3, and each group includes only 1 user data. The three groups are respectively recorded as group 1, group 2 and group 3, where group 3 is the group corresponding to the most recent time period (recorded as time period 1), group 2 is the group corresponding to the previous time period of time period 1 (recorded as time period 2), and group 1 is the group corresponding to the previous time period of time period 2 (recorded as time period 3). The user data in group 1 is recorded as user data 1, the user data in group 2 is recorded as user data 2, and the user data in group 3 is recorded as user data 3. It can be understood that the actual application is not limited to this. For example Figure 6As shown, the number of visits to a specific web page (assuming they are 5 times, 2 times, and 0 times, respectively) can be extracted from user data 1 in group 1, user data 2 in group 2, and user data 3 in group 3, respectively. Then, it can be determined that the number of visits to the specific web page by potential user 30 is on an upward trend, and thus it can be determined that the changing trend of the potential user 30's willingness to accept the target service is also on an upward trend.

[0065] The above embodiment analyzes the changing trend of the potential user 30's willingness to accept the target service, so as to determine whether the potential user 30's willingness to accept the target service is gradually increasing, and finally determine whether the potential user 30 is likely to accept the target service, so as to facilitate the customer to judge whether it is necessary to continue to follow up with the potential user 30 and whether the follow-up strategy needs to be adjusted.

[0066] In some embodiments, the potential user 30 may also be scored for his / her willingness to accept the target service based on the description information corresponding to the potential user 30, and a score of the potential user 30 may be obtained, and the score is positively correlated with the potential user 30's willingness to accept the target service. The target category to which the potential user 30 belongs may be determined from multiple candidate categories based on the above score, wherein the multiple candidate categories correspond to multiple score intervals respectively, and the score of the potential user 30 is within the score interval corresponding to the target category.

[0067] In this embodiment, by classifying the potential users 30, clearer, more explicit, and more intuitive category information can be provided to describe the potential users 30's willingness to accept the target service. For example, the willingness of the potential users 30 can be divided into five candidate categories, namely low willingness (score less than 30 points), medium-low willingness (score between 30 and 50 points), medium willingness (score between 50 and 70 points), medium-high willingness (score between 70 and 80 points), and high willingness (score above 80 points), which represent the potential users 30's acceptance of the target service from low to high. If the potential user 30 scores 75 points, it means that the potential user 30 belongs to the medium-high willingness category, that is, has a high willingness to accept the target service. It can be understood that the number of categories and the score intervals corresponding to each category are only exemplary and are not intended to limit the present application.

[0068] In some embodiments, the number of potential users 30 is greater than 1. The scores of the multiple potential users 30 may be obtained respectively, and the multiple potential users 30 may be ranked based on the scores of the multiple potential users 30. In this way, the ranking of the willingness of each potential user 30 to accept the target service may be determined, so that it is convenient to arrange the members of the customer service team to focus their main efforts on following up the potential users 30 with higher scores.

[0069] In some embodiments, the feature tag may not be easy to understand, or there may be ambiguity. Therefore, the target feature tag of at least one feature tag may be pre-associated with pre-generated explanation information, which is used to explain the meaning of the associated target feature tag. Among them, the target feature tag may include some or all feature tags in at least one feature tag. For example, the target feature tag includes "enterprise status", and the associated explanation information may be "enterprise status represents the overall operating status and health of the enterprise at a certain moment or in a certain period of time". On this basis, the explanation information associated with the target feature tag can be obtained, and the description information corresponding to the potential user 30 is generated based on the explanation information associated with the target feature tag and the feature data corresponding to the target feature tag. Among them, the description information may only include the feature data corresponding to the target feature tag and the explanation information associated with the target feature tag, or may include the target feature tag, the feature data corresponding to the target feature tag, and the explanation information associated with the target feature tag. Continuing with the above example, the description information may be "the overall operating status and health of the enterprise in the past year are good", or "the enterprise status (overall operating status and health) of the enterprise in the past year is good". By associating the feature tag with the explanation information and referring to the explanation information when generating the description information, the generated description information can be made easier to understand and the ambiguity of the description information can be reduced, thereby facilitating the reading and viewing of members of the customer service team. Furthermore, for feature tags that are not associated with the explanation information, the description information can be generated based only on the feature tag itself and the feature data corresponding to the feature tag.

[0070] In some embodiments, the description information is information in a specified format. The specified format may be, for example, JSON format. In order to ensure that the description information is output in the correct format, a format check may be performed on the description information to determine whether the description information is information in the specified format. If the check result indicates that the description information is not information in the specified format, the description information may be regenerated so as to adjust the description information to the specified format.

[0071] In some embodiments, a follow-up suggestion for the potential user 30 can also be generated based on the description information through a large language model, and the follow-up suggestion is used to describe a set of means to improve the potential user 30's willingness to accept the target service. For example, the description information may include the potential user 30's concerns about the target service, such as price, logistics time, etc. Follow-up suggestions for the potential user 30 can be generated based on the above concerns. For example, if the description information includes "the customer thinks the product price is too high" or "the logistics time is too long", the follow-up suggestion may include reducing the product price or reducing the logistics time.

[0072] In some embodiments, after the description information is generated, the description information can also be displayed on the front end. Further, in the example where other information (such as follow-up suggestions, scores, target categories) is generated, the other information and the description information can also be displayed on the front end together.

[0073] The embodiments of the present application can be applied to e-commerce platforms, such as cross-border e-commerce platforms. Salespeople of cross-border e-commerce platforms need to follow up on potential users who may enter the cross-border e-commerce platforms. However, salespeople may face the following problems:

[0074] (1) New sales challenges: New salespeople may face pressure to quickly learn and adapt to sales processes. They need to quickly master product knowledge, market trends, and customer communication skills in order to conduct sales activities effectively.

[0075] (2) Time-consuming customer inventory: In the traditional sales process, collecting and analyzing user data can be a time-consuming and tedious process. The sales team needs to integrate information from multiple sources (such as CRM systems, social media, market research reports, etc.) in order to fully understand the needs and preferences of potential users.

[0076] (3) Insufficient experience: Due to lack of industry experience and in-depth understanding of user behavior, new salespeople may find it difficult to accurately judge users’ purchasing intentions and potential needs, which may lead to missed sales opportunities or incorrect sales strategies.

[0077] (4) Easy to misclassify: When dealing with large amounts of user data, salespeople may incorrectly classify or evaluate user categories due to human factors, such as misjudging potential large customers as low-value customers or failing to identify real sales barriers.

[0078] In view of these background issues, this application aims to automate and optimize the sales process by introducing artificial intelligence technology. This technology can help salespeople quickly gain user insights, reduce the time of manual data collation, and provide data-based follow-up suggestions, thereby improving sales efficiency and success rate. At the same time, by using large language model analysis and machine learning algorithms, it can reduce human errors, ensure accurate classification and evaluation of user data, and help salespeople better understand and meet user needs.

[0079] At present, the user insight function in the relevant technology cannot be analyzed according to the professional terms of commodities and industries, the generated results are not professional enough, and no complete solution is provided from the professional knowledge base to user analysis and action suggestions. This relatively shallow intention judgment and analysis summary is difficult to help experienced sales, and may even lead to identification errors. In addition, at the level of overall solutions, there is currently no process-based solution from time-series heterogeneous raw data to user insights in the industry, but users are required to process it themselves. The solution of this application will further extend the capability boundary, touching the user's original data, and will be more automated and intelligent. The solution of this application can not only analyze a single user, but also sort all contacted users to identify potential users and facilitate energy arrangement.

[0080] In terms of core technology, most related technologies use the BERT model to achieve functions such as intention classification and key information extraction. The one-stop user analysis software of this application can use the GPT4 generative large language model. Compared with the BERT bidirectional encoding model, through prompt engineering and knowledge injection, this application can obtain more natural and professional insight results. Through the chain of thinking, it can comprehensively consider a variety of information and complete multiple tasks. In actual applications, the output content of the solution of this application has received a higher professional score.

[0081] Combine the following Figure 7 The system architecture diagram shown and Figure 8 The overall flow chart shown illustrates the overall process of the embodiment of the present application. The overall process of the embodiment of the present application mainly includes processes such as data collection, intelligent processing, format verification and front-end display.

[0082] Data collection

[0083] The collected data includes business information, communication data, behavior data, etc. Among them, business information includes but is not limited to the name, scale, business scope, person in charge (KP), telephone number, address, status, e-commerce experience, etc. of the enterprise user. Communication data includes but is not limited to telephone communication data, SMS communication data, door-to-door communication data, email communication data, etc. Behavior data includes but is not limited to users' content browsing, website operations, participation in activities, participation in exhibitions, etc. Data collection methods include but are not limited to at least one of the following:

[0084] Through SQL queries, key user data can be extracted from the e-commerce platform’s database. This process is automated, ensuring seamless integration of all relevant information from order history to user interactions, providing real-time and accurate data flow to the analysis engine.

[0085] Web trackers are used to automatically collect key user data from various websites on the Internet. These websites include official websites, recruitment platforms, business information websites, and industry-related professional websites. This method can be used to discover potential users' business activities, expansion plans, industry trends, and human resource changes. After obtaining this information, the system will analyze the data to reveal market opportunities and potential business partnerships, allowing the sales team to more accurately target users and optimize sales strategies.

[0086] Robotic process automation (RPA) technology plays a key role in the application scenarios of mobile applications. By simulating common interactive behaviors of users in various mobile applications, such as adding friends, sending files, browsing user sharing and publishing content, RPA technology can efficiently identify and record user activity and content preferences.

[0087] The system is set up with smart website hooks that are activated when specific actions occur on the internal website. This means that any sales activity or customer feedback is automatically recorded and included in the analysis, ensuring that the decision-making process has the latest information.

[0088] In the system of this application, the data manually entered by CRM plays a supplementary role. The manually recorded information can capture details that may be missed by the automated system, such as personalized user communication and specific sales opportunities. This data helps to build a comprehensive customer portrait and enhance sales strategies.

[0089] Intelligent processing

[0090] In the "intelligent processing" module of the intelligent information processing system, a series of advanced artificial intelligence models and machine learning algorithms are applied to transform raw data into valuable insights.

[0091] (1) Sort by time. To ensure the time sequence of data analysis, the collected user data is first sorted by time. This step helps capture the time clues of sales activities, thereby analyzing trend changes and behavior patterns.

[0092] (2) Aggregate and organize by cycle and operation. The data is grouped into different cycle and operation categories through aggregation processing to enable more detailed analysis of sales activities. For example, business performance can be evaluated by week, month, or specific sales stage.

[0093] (3) Large language model extracts key information from the data based on preset label requirements, such as "Extract [name]". This process involves identifying and extracting the data points that have the greatest impact on sales progress.

[0094] (4) The large language model classifies the data based on the labels. Next, the same large language model is used to classify the data. Each data point is classified into the corresponding category based on its label for subsequent data analysis and application.

[0095] (5) In this step, the large language model is used to rewrite and optimize the data expression. In this way, the large language model not only improves the readability of the data, but also generates clearer reports and summaries (i.e., descriptive information). For example, the original data is: Enter the keyword xxx, which will be rewritten in combination with other information during this period as: During this period, users often operate on the platform official website, search for xxx content, and express their willingness to learn more on the phone.

[0096] (6) Obtaining explanations of professional terms based on tags, i.e., knowledge injection. To ensure the professionalism and accuracy of the data, the system obtains and attaches definitions and explanations of professional terms based on relevant tags. This process is crucial for non-professional users to understand complex data.

[0097] (7) Summary generation. Finally, the summary generated by the large language model can provide a concise and comprehensive overview of the entire data set. This step is particularly important for salespeople, allowing them to quickly grasp key information and insights.

[0098] Through these carefully designed steps, the intelligent processing module of the present application can process complex raw data into strategically valuable business intelligence, providing a solid foundation for making wise sales decisions.

[0099] Format Validation

[0100] (1) The system first uses regular expressions, a powerful text matching tool, to accurately verify the format of the summaries generated by the large language model. Regular expressions can define complex matching rules to check for specific patterns in the summaries, such as date formats, numeric sequences, email addresses, or custom text structures. If the verification process finds that the summary does not match the established regular expression pattern, the system will mark these anomalies and trigger the regeneration process until the generated text fully meets the preset format requirements.

[0101] (2) After completing the format verification, the system will add front-end style tags to the summary content. These tags are used to define the display style of the text, including font, color, bold, underline, etc. This not only improves the visual appeal of the data presentation, but also enhances the readability of the data and user experience.

[0102] (3) Finally, the system converts the validated and styled text into JSON format so that it can be seamlessly connected to the front-end application. JSON format data supports nested structures, is suitable for the representation of complex data, and is easy for front-end technology to parse and use.

[0103] Through this series of steps in the "Format Validation and Conversion" module, the system ensures that the summary is not only accurate in content, but also in a standardized format and easy to display and integrate in various front-end environments. This module is a key link in ensuring the quality of system output and is essential for providing a high-quality user experience.

[0104] Front-end display

[0105] The above information is sent from the server to the front-end page through the network protocol to display the user's willingness score, interaction summary, focus, and abnormal situations, and users with higher willingness scores are displayed in front of the front-end page.

[0106] The embodiment of the present application also proposes a method for self-generating prompt words in order to improve the indicators of natural language understanding tasks. The embodiment of the present application pioneered the specification of heterogeneous time series data collation, mining of feature labels in the field, generation of prompt words, and finally a professional sales auxiliary link after content verification and style modification, which provides an intelligent solution for traditional sales work, greatly shortens the sales cycle of products and improves customer satisfaction.

[0107] See also Fig. 9 , the embodiment of the present application also provides a method for generating a summary, the method comprising:

[0108] Step S22: obtaining user data of potential users of the e-commerce platform collected within a historical period;

[0109] Step S24: using the user data as input data of the large language model, and using at least one pre-generated feature tag as prompt information of the large language model, so as to extract feature data corresponding to the at least one feature tag from the user data through the large language model; the at least one feature tag is related to the potential user's willingness to accept the e-commerce service provided by the e-commerce platform;

[0110] Step S26: Generate a summary corresponding to the potential user based on the at least one feature tag and its corresponding feature data, wherein the summary is used to describe the potential user's willingness to accept the e-commerce service.

[0111] The e-commerce platform of the embodiment of the present application corresponds to the service platform in the embodiment of the aforementioned data processing method, and the e-commerce service of the embodiment of the present application corresponds to the target service in the embodiment of the aforementioned data processing method. The specific implementation details of this embodiment are detailed in the embodiment of the aforementioned data processing method, which will not be repeated here.

[0112] An embodiment of the present application further provides a computer device, which comprises at least a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any of the foregoing embodiments is implemented.

[0113] Fig.10 A more specific hardware structure diagram of a computer device provided in an embodiment of the present application is shown, and the device may include: a processor 302, a memory 304, an input / output interface 306, a communication interface 308, and a bus 310. The processor 302, the memory 304, the input / output interface 306, and the communication interface 308 are connected to each other in communication within the device through the bus 310.

[0114] The processor 302 can be implemented by a general central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), or one or more integrated circuits, etc., for executing relevant programs to implement the technical solutions provided in the embodiments of the present application. The processor 302 can also include a graphics card, which can be a Nvidia titan X graphics card or a 1080Ti graphics card, etc.

[0115] The memory 304 can be implemented in the form of a read-only memory (ROM), a random access memory (RAM), a static storage device, a dynamic storage device, etc. The memory 304 can store an operating system and other application programs. When the technical solution provided in the embodiment of the present application is implemented by software or firmware, the relevant program code is stored in the memory 304 and called and executed by the processor 302.

[0116] The input / output interface 306 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0117] The communication interface 308 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0118] The bus 310 comprises a pathway for transmitting information between the various components of the device, such as the processor 302 , the memory 304 , the input / output interface 306 , and the communication interface 308 .

[0119] It should be noted that, although the above device only shows the processor 302, the memory 304, the input / output interface 306, the communication interface 308 and the bus 310, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiment of the present application, and does not necessarily include all the components shown in the figure.

[0120] An embodiment of the present application provides a computer program product, including a computer program, which implements the method described in any embodiment of the present application when executed by a processor.

[0121] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the foregoing embodiments.

[0122] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computer device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0123] Each embodiment in the present application is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts refer to the partial description of the method embodiment. The device embodiment described above is merely schematic, wherein the modules described as separate components may or may not be physically separated, and the functions of each module can be implemented in the same one or more software and / or hardware when implementing the embodiment scheme of the present application. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the embodiment scheme. Those of ordinary skill in the art can understand and implement it without paying creative work.

[0124] The above is only a specific implementation of the embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the embodiment of the present application. These improvements and modifications should also be regarded as the protection scope of the embodiment of the present application.

Claims

1. A data processing method, the method comprising: Obtain user data of potential users of the service platform collected within a historical time period; The user data is used as input data of the large language model, and at least one pre-generated feature tag is used as prompt information of the large language model, so as to extract feature data corresponding to the at least one feature tag from the user data through the large language model; the at least one feature tag is related to the potential user's willingness to accept the target service provided by the service platform; Descriptive information corresponding to the potential user is generated according to the at least one feature tag and its corresponding feature data, wherein the descriptive information is used to describe the potential user's willingness to accept the target service.

2. According to the method of claim 1, the amount of the user data is greater than 1; the generating the description information corresponding to the potential user according to the at least one feature tag and its corresponding feature data respectively comprises: Acquire at least one group of user data, wherein any group includes at least one piece of user data, and each piece of user data in the same group is user data generated within the same time period; For any one of the at least one group, determining description information of the potential user in a time period corresponding to the group according to the at least one feature tag and feature data corresponding to the at least one feature tag respectively extracted from user data of the group; Based on the description information of the potential user in the time period respectively corresponding to the at least one group, a change trend of the potential user's willingness to accept the target service is determined.

3. The method according to claim 2, wherein the at least one feature tag comprises a user behavior tag, wherein the user behavior tag is used to describe an operation performed by the potential user on an object related to the target service; The determining, based on the description information of the potential user in the time period respectively corresponding to the at least one group, a change trend of the potential user's willingness to accept the target service includes: For any one of the at least one group, obtaining the number of feature data corresponding to a specified user behavior tag extracted from user data of the group; Based on the quantity of feature data corresponding to the designated user behavior tag extracted from the user data of each group, a change trend of the potential user's willingness to accept the target service is determined.

4. The method according to claim 2, wherein the user data comprises data collected from multiple data sources; The obtaining of at least one group of user data comprises: Acquire time information corresponding to the user data collected from the multiple data sources; Performing time alignment on the user data collected from the multiple data sources based on the time information; The time-aligned user data is grouped to obtain at least one group of user data.

5. The method according to claim 1, further comprising: Scoring the potential user's willingness to accept the target service according to the description information corresponding to the potential user, and obtaining a score of the potential user, wherein the score is positively correlated with the potential user's willingness to accept the target service; A target category to which the potential user belongs is determined from a plurality of candidate categories based on the score, wherein the plurality of candidate categories correspond to a plurality of score intervals respectively, and the score of the potential user is within the score interval corresponding to the target category.

6. The method according to claim 1, wherein the target feature tag of the at least one feature tag is associated with pre-generated explanation information, and the explanation information is used to explain the meaning of the associated target feature tag; and the generating of the description information corresponding to the potential user according to the at least one feature tag and its corresponding feature data comprises: Obtaining explanation information associated with the target feature tag; Descriptive information corresponding to the potential user is generated based on the explanation information associated with the target feature tag and the feature data corresponding to the target feature tag.

7. The method according to claim 1, wherein the description information is information in a specified format; the method further comprises: Performing a format check on the description information to determine whether the description information is in the specified format; If the verification result indicates that the description information is not information in the specified format, the description information is regenerated.

8. The method according to claim 1, wherein the user data of the potential user comprises at least one of the following: User data recorded in the database of the service platform; User data of the potential user collected through web page trackers on the Internet; data obtained by simulating the behavior of the potential user through robotic process automation; Data generated when the potential user performs actions on a website related to the target service; Data manually recorded during interviews with the potential users.

9. The method according to claim 1, further comprising: A follow-up suggestion for the potential user is generated according to the description information, where the follow-up suggestion is used to describe a set of means for improving the potential user's willingness to accept the target service.

10. According to the method of claim 1, the at least one feature label is obtained through a pre-trained label generation model.

11. The method according to claim 10, further comprising: Obtaining sample data and sample labels corresponding to the sample data; The sample label is used to indicate whether the user to whom the sample data belongs is willing to accept the target service; Obtaining a current prompt word output by the label generation model based on the sample data and its corresponding sample label, wherein the current prompt word includes at least one current predicted label; Determining a current performance indicator corresponding to the current prompt word, wherein the current performance indicator is used to evaluate the accuracy of the at least one current predicted label; If the difference between the current performance indicator and the historical performance indicator is greater than a preset threshold, return to the step of obtaining the current prompt word output by the label generation model based on the sample data, wherein the historical performance indicator is the performance indicator corresponding to the previous prompt word output by the label generation model.

12. The method according to claim 11, further comprising: If the difference between the current performance indicator and the historical performance indicator is not greater than a preset threshold, the at least one current prediction label is determined as the at least one feature label.

13. The method according to claim 12, wherein if the difference between the current performance indicator and the historical performance indicator is not greater than a preset threshold, determining the at least one current prediction label as the at least one feature label comprises: If the difference between the current performance index and the historical performance index is not greater than a preset threshold, obtaining the label stability of the sample data, wherein the label stability is used to evaluate the consistency of the predicted labels output by the label generation model for the sample data multiple times; If the label stability is higher than a preset stability threshold, the at least one current predicted label is determined as the at least one feature label.

14. The method according to claim 13, further comprising: If the label stability is not higher than a preset stability threshold, the sample label is adjusted.

15. A method for generating a summary, the method comprising: Obtain user data of potential users of the e-commerce platform collected within a historical period; The user data is used as input data of the large language model, and at least one pre-generated feature tag is used as prompt information of the large language model, so as to extract feature data corresponding to the at least one feature tag from the user data through the large language model; the at least one feature tag is related to the potential user's willingness to accept the e-commerce service provided by the e-commerce platform; A summary corresponding to the potential user is generated according to the at least one feature tag and its corresponding feature data, wherein the summary is used to describe the potential user's willingness to accept the e-commerce service.

16. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 15 is implemented.

17. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 15 when executing the program.

18. A computer program product, comprising a computer program, which implements the method according to any one of claims 1 to 15 when executed by a processor.

Images