User quantification method and device based on dialogue text embedding characteristics, medium and program product

Abstract

The embodiment of the invention provides a user quantification method and device based on dialogue text embedding characteristics, a medium and a program product, and relates to the technical field of user quantification. The method comprises the steps of obtaining user portrait features and telemarketing dialogue text of a to-be-quantized user; extracting a plurality of key text fragments from the telemarketing dialogue text based on a preset process node division rule, and respectively obtaining an embedding feature of each key text fragment; performing weighted aggregation on each embedding feature based on preset weight information to obtain a dialogue embedding feature; splicing the user portrait features and the dialogue embedding features to obtain a comprehensive feature set; and determining a quantification result corresponding to the to-be-quantified user based on the comprehensive feature set by using a preset user quantification model. According to the embodiment of the invention, the dialogue text is subjected to fragment segmentation according to the process nodes, and the dialogue features are determined according to the weight of each fragment and the extracted embedded features, so that the accuracy and comprehensiveness of dialogue feature extraction are improved, and the user quantification accuracy is effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of user quantification, and in particular to a user quantification method, device, medium, and program product based on conversation text embedding features. Background Art

[0002] In telemarketing scenarios, user quantification is key to optimizing communication strategies and improving conversion rates. Traditional quantification models rely primarily on user profile features (such as age, gender, and purchase history) and conversation tags. These tags are generated based on telemarketing conversation text using tag generation algorithms.

[0003] However, since the label generation algorithm may not accurately capture the complete context or intent of the conversation, resulting in incorrect labels or information loss, directly using these conversation labels for user quantification lacks flexibility, resulting in low accuracy of user quantification. Summary of the Invention

[0004] The purpose of the embodiments of the present application is to provide a user quantification method, device, medium and program product based on conversation text embedding features to improve the accuracy of user quantification.

[0005] In a first aspect, an embodiment of the present application provides a user quantification method based on conversation text embedding features, comprising: Obtain user portrait features and telemarketing conversation texts of the users to be quantified; Extracting multiple key text segments from the telemarketing conversation text based on a preset process node division rule, and obtaining embedding features of each key text segment; Performing weighted aggregation on each of the embedding features based on preset weight information to obtain a conversation embedding feature; Concatenating the user portrait features and the conversation embedding features to obtain a comprehensive feature set; A preset user quantization model is used to determine a quantization result corresponding to the user to be quantified based on the comprehensive feature set.

[0006] In an embodiment of the present application, by segmenting the conversation text into segments according to process nodes and determining the conversation features based on the weights of each segment and the extracted embedded features, the accuracy and comprehensiveness of conversation feature extraction are improved, thereby effectively improving user quantification accuracy.

[0007] In some possible embodiments, performing weighted aggregation on the embedded features based on preset weight information to obtain the conversation embedded features includes: Acquiring attribute information of each key text segment, and determining weight information corresponding to each key text segment according to the attribute information of each key text segment; The embedded features are weighted and aggregated based on the weight information to obtain a conversation embedded feature.

[0008] In the embodiment of the present application, by obtaining the attribute information of each text segment to determine the corresponding weight, the accuracy of the integration of dialogue text features is further improved, thereby effectively improving the user quantification accuracy.

[0009] In some possible embodiments, obtaining the attribute information of each key text segment and determining the weight information corresponding to each key text segment according to the attribute information of each key text segment includes: Acquiring attribute information of each of the key text segments; wherein the attribute information includes at least one of a text length attribute, a dialogue clarity attribute, a dialogue efficiency attribute, and a user intent clarity attribute; The importance of each key text segment is determined based on the attribute information of each key text segment, and weight information corresponding to each key text segment is determined according to the importance of each key text segment.

[0010] In an embodiment of the present application, by comprehensively determining the importance of a text segment based on multiple attribute information of the segment and determining the integration weight of each text segment according to the importance, the accuracy of the integration of dialogue text features is further improved, thereby effectively improving the user quantification accuracy.

[0011] In some possible embodiments, determining a quantization result corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model includes: Determining a quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model; When the quantization score is greater than a preset score threshold, the user to be quantified is identified as a target user.

[0012] In the embodiment of the present application, the quantization result of the user is determined by comparing the quantization score with a preset threshold, thereby further improving the flexibility of user quantization.

[0013] In some possible embodiments, determining a quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model includes: Determine a first target quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset first user quantization model, and determine a second target quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset second user quantization model; A quantization score corresponding to the to-be-quantized user is determined based on the first target quantization score and the second target quantization score.

[0014] In an embodiment of the present application, a multi-objective quantization score of a user is determined by a multi-objective quantization model, and a comprehensive quantization result is determined based on the multi-objective quantization score, thereby further improving the flexibility of user quantization.

[0015] In some possible embodiments, the user quantification method based on conversation text embedding features further includes: Add the user object identified as the target user to the target user list; Counting the actual telesales conversion rate corresponding to the target user list according to a preset statistical period; Based on the comparison between the actual telesales conversion rate and the preset conversion rate threshold, the preset score threshold is adjusted.

[0016] In the embodiment of the present application, by collecting statistics on the actual conversion rate of target users on a periodic basis and dynamically adjusting the score threshold for identifying target users based on the comparison between the conversion rate and the preset threshold, the flexibility and accuracy of user quantification are further improved.

[0017] In some possible embodiments, determining a quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model includes: Determining a first quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model; Generate a label feature corresponding to the user to be quantified based on the telemarketing conversation text using a preset label generation algorithm; Concatenate the user portrait features and the label features to obtain a second comprehensive feature set; Determining a second quantization score corresponding to the user to be quantized based on the second comprehensive feature set using the user quantization model; A quantization score corresponding to the to-be-quantized user is determined based on the first quantization score and the second quantization score.

[0018] In an embodiment of the present application, the quantization scores are determined respectively by using segmented fusion embedding features based on the conversation text and label features based on the conversation text, and the user's quantization result is determined comprehensively based on the two quantization scores, thereby further improving the accuracy of user quantization.

[0019] In a second aspect, an embodiment of the present application provides a user quantification device based on conversation text embedding features, comprising: The data acquisition module is used to obtain the user portrait features and telemarketing conversation texts of the users to be quantified; A feature extraction module, configured to extract a plurality of key text segments from the telemarketing conversation text based on a preset process node division rule, and obtain embedded features of each of the key text segments; A feature aggregation module, configured to perform weighted aggregation on each of the embedding features based on preset weight information to obtain a conversation embedding feature; A feature splicing module, configured to splice the user portrait features and the conversation embedding features to obtain a comprehensive feature set; The user quantization module is configured to determine a quantization result corresponding to the user to be quantified based on the comprehensive feature set by using a preset user quantization model.

[0020] In a third aspect, an embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor can implement the method described in any embodiment of the first aspect when executing the program.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program, wherein when the computer program is executed by a processor, it can implement the method described in any embodiment of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A flowchart of a user quantification method based on conversation text embedding features provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a user quantification device based on conversation text embedding features provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0026] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0027] like Figure 1 As shown, the embodiment of the present application provides a user quantification method based on conversation text embedding features, which may include the following steps: S1. Obtain the user portrait features and telemarketing conversation text of the user to be quantified.

[0028] Specifically, the user to be quantified is the user currently requiring quantification. First, basic data corresponding to the user is obtained, including the user's user profile features and telemarketing conversation text. User profile features include static data such as the user's age, gender, and purchase history. Telemarketing conversation text is obtained by converting the telemarketing system's telephone sales conversations (voice calls) with the user into text.

[0029] For example, before step S1, the user metadata (including call duration, call time, number of connections, etc.) of each user can be obtained to perform a preliminary screening of users to obtain users to be quantified. For example, if a user does not respond, effective communication is not established with the user, or there is no effective response information, these users are excluded from consideration in the user quantification process.

[0030] S2. Extract multiple key text fragments from the telemarketing conversation text based on the preset process node division rules, and obtain the embedded features of each key text fragment respectively.

[0031] Specifically, the process node division rules can divide telemarketing conversation texts into different process nodes. For example, based on the pre-set keywords of the AI ​​robot (the interlocutor of the telemarketing system), process nodes such as the opening node, demand generation node, objection handling node, and closing node in the telemarketing conversation text can be identified, thereby dividing the telemarketing conversation text into the opening stage, demand generation stage, objection handling stage, and closing stage.

[0032] Then, according to the stages into which the telemarketing conversation text is divided, multiple key text fragments with high reference significance for user quantification can be extracted from it. For example, the conversation texts in the demand mining stage and the objection handling stage can be extracted as key text fragments respectively.

[0033] For the multiple key text fragments extracted, embedding features can be extracted for each of them. For example, the embedding features of different key text fragments can be extracted based on the same feature extraction method, or different feature extraction methods can be set to extract the embedding features of the key text fragments in the corresponding dialogue stages. Among them, the embedding features of the dialogue text can be extracted based on the large open source language model on the market. It should be noted that feature embedding is a technology that converts high-dimensional sparse features into low-dimensional dense vectors. It is particularly suitable for processing high-dimensional data, such as text, images or graph structured data. Through the embedding method, sparse high-dimensional features are compressed into a low-dimensional continuous vector space, so that these features can be better represented and processed in the machine learning model.

[0034] S3. Perform weighted aggregation on each embedded feature based on the preset weight information to obtain the conversation embedded feature.

[0035] Specifically, corresponding weights can be set for different dialogue stages in the telemarketing dialogue text. That is, different key text fragments are set with corresponding weights. Based on these preset weight information, each embedded feature can be weighted and aggregated.

[0036] Exemplarily, the preset weight information may be: a weight of 0.6 configured for the demand mining stage and a weight of 0.4 configured for the objection handling stage; assuming that based on the current telemarketing conversation text, the embedded feature corresponding to the key text fragment extracted in the demand mining stage is E1, and the embedded feature corresponding to the key text fragment extracted in the objection handling stage is E2, then the conversation embedding feature obtained by weighted aggregation is: E=0.6*E1+0.4*E2.

[0037] S4. Concatenate user portrait features and conversation embedding features to obtain a comprehensive feature set.

[0038] For example, a feature selection algorithm based on interactive information can be used to eliminate redundant features from the extracted features; then, the high-dimensional dense embedding features and the low-dimensional sparse portrait features are aligned to a unified feature space through the model adaptive mapping layer, that is, the user portrait features and the conversation embedding features are spliced ​​to form a comprehensive feature set.

[0039] S5. Determine a quantization result corresponding to the user to be quantified based on the comprehensive feature set using a preset user quantization model.

[0040] Specifically, based on the obtained comprehensive feature set, it can be input into a preset user quantization model to obtain a quantization result corresponding to the user to be quantized output by the model.

[0041] For example, a machine learning model (such as a neural network or gradient boosting decision tree) can be used to construct an initial model. This initial model is then trained based on the collected training dataset to obtain a corresponding user quantification model. For example, the training data can be data consisting of several sample features and their corresponding user quantification labeling results.

[0042] Based on this, by segmenting the conversation text into segments according to process nodes and determining the conversation features based on the weights of each segment and the extracted embedded features, the accuracy and comprehensiveness of conversation feature extraction are improved, thereby effectively improving user quantification accuracy.

[0043] In some possible embodiments, step S3, performing weighted aggregation on each embedded feature based on preset weight information to obtain a conversation embedded feature, may include: S301, obtaining attribute information of each key text segment, and determining weight information corresponding to each key text segment according to the attribute information of each key text segment; S302: Perform weighted aggregation on each embedded feature based on the weight information to obtain a conversation embedded feature.

[0044] It should be noted that the telemarketing conversation texts generated by different users are different, and the text content and attributes corresponding to multiple key text fragments in the conversation text generated by the same user are also different.

[0045] Therefore, the weight information corresponding to each key text segment can be determined based on the attribute information of each key text segment in the conversation text generated by the current user.

[0046] Based on this, by obtaining the attribute information of each text segment to determine the corresponding weight, the accuracy of dialogue text feature integration is further improved, thereby effectively improving user quantification accuracy.

[0047] In some possible embodiments, step S301, obtaining attribute information of each key text segment and determining weight information corresponding to each key text segment based on the attribute information of each key text segment, may include: S3011. Acquire attribute information of each key text segment; wherein the attribute information includes at least one of a text length attribute, a dialogue clarity attribute, a dialogue efficiency attribute, and a user intent clarity attribute; S3012: Determine the importance of each key text segment based on the attribute information of each key text segment, and determine weight information corresponding to each key text segment according to the importance of each key text segment.

[0048] Exemplarily, the attribute information includes one or more of a text length attribute, a conversation clarity attribute, a conversation efficiency attribute, and a user intention clarity attribute.

[0049] Among them, the text length attribute is used to characterize the text length (such as the number of characters) of the key text fragment. For example, the longer the text length is, the deeper the user conversation turn corresponding to the key text fragment is (the more conversation content), and the more important the corresponding feature is to be included in the user quantification, and vice versa.

[0050] The conversation clarity attribute characterizes the clarity of the key text segment. For example, factors such as poor signal quality, noisy environments, or accents can lead to poor clarity. These factors can be captured based on the specific call situation and converted into corresponding conversation clarity attributes according to pre-set rules. It's understood that higher conversation clarity indicates more reliable content within the key text segment, and therefore, its corresponding feature is more important for user quantification, and vice versa.

[0051] The conversation efficiency attribute is used to characterize the timeliness of feedback on the user conversation corresponding to the key text fragment, and is mainly determined based on the waiting time from the AI ​​robot asking a question to the user's response (the longer the waiting time, the lower the conversation efficiency). It can be understood that the higher the conversation efficiency, the higher the clarity of the user conversation corresponding to the key text fragment, and the more important it is to include its corresponding features in user quantification. Conversely, the lower the conversation efficiency, the more hesitant the user conversation process corresponding to the key text fragment, and the less important it is to include its corresponding features in user quantification.

[0052] The user intent clarity attribute characterizes the clarity of the user's intent in the conversation corresponding to the key text segment, specifically, the degree to which the user clearly expressed affirmative or negative intent. For example, keywords (such as "confirmed," "uncertain," and "don't know") can be set to reflect the clarity of the user's intent. The clarity of the user's intent in the conversation feedback can be measured based on the number of keyword hits in the actual conversation. It is understood that a higher degree of user intent clarity indicates a more explicit conversation corresponding to the key text segment, and thus the corresponding feature is more important for inclusion in user quantification. The same applies vice versa.

[0053] Exemplarily, for each key text fragment, the attribute scores corresponding to different attribute information (text length attribute, dialogue clarity attribute, dialogue efficiency attribute and user intention clarity attribute) can be calculated respectively according to preset rules, and the comprehensive attribute score corresponding to the key text fragment can be determined. Then, the importance of each key text fragment can be measured according to the comprehensive attribute score corresponding to the key text fragment, so as to determine the weight information corresponding to each key text fragment according to the importance of each key text fragment (the higher the importance, the higher the weight).

[0054] Based on this, by comprehensively determining the importance of the text fragment based on the various attribute information of the text fragment and determining the integration weight of each text fragment according to the importance, the accuracy of the integration of dialogue text features is further improved, thereby effectively improving the user quantification accuracy.

[0055] In some possible embodiments, step S5, determining a quantization result corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model, may include: S501: Determine a quantization score corresponding to the user to be quantized based on a comprehensive feature set using a preset user quantization model; S502: When the quantization score is greater than a preset score threshold, identify the user to be quantified as a target user.

[0056] It should be noted that, by using the preset user quantitative model, a corresponding quantitative score can be output based on the comprehensive feature set of the input model. The quantitative score represents the potential degree of the user relative to the preset quantitative target, such as purchase propensity, complaint propensity, etc.

[0057] Then, through a preset score threshold (which can be set according to actual needs and experimental tests), target users whose quantitative scores are higher than the score threshold can be identified and screened as the users to be focused on in the future.

[0058] Based on this, the user's quantification result is determined by comparing the quantization score with a preset threshold, thereby further improving the flexibility of user quantification.

[0059] In some possible embodiments, step S501, determining a quantization score corresponding to the user to be quantized based on a comprehensive feature set using a preset user quantization model, may include: S5011. Determine a first target quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset first user quantization model, and determine a second target quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset second user quantization model; S5012: Determine a quantization score corresponding to the user to be quantized based on the first target quantization score and the second target quantization score.

[0060] It should be noted that in addition to single-target user quantification, multi-target quantification strategies can also be set according to needs. For example, users with high purchase propensity and low complaint propensity are identified, and corresponding user quantification strategies can be set based on purchase propensity and complaint propensity respectively.

[0061] Specifically, different user quantification models can be set based on different user quantification goals. For example, a first user quantification model can be used to quantify a user's purchase propensity to obtain a first target quantification score, and a second user quantification model can be used to quantify a user's complaint propensity to obtain a corresponding second target quantification score.

[0062] Then, the final quantization score is determined by comprehensively considering the first target quantization score and the second target quantization score. For example, the quantization score of the user to be quantized can be obtained by weighted summing the first target quantization score and the second target quantization score based on a preset weight.

[0063] Exemplarily, a preset multi-objective user quantification model may be used to process the user's comprehensive feature set at one time to obtain a user quantification score that can meet multiple objectives.

[0064] Based on this, the user's multi-objective quantization score is determined by a multi-objective quantization model, and the comprehensive quantization result is determined based on the multi-objective quantization score, thereby further improving the flexibility of user quantization.

[0065] In some possible embodiments, the user quantification method based on conversation text embedding features may further include: S601, adding the user object identified as the target user to the target user list; S602: Calculate the actual telemarketing conversion rate corresponding to the target user list according to a preset statistical period; S603: Based on the comparison between the actual telesales conversion rate and the preset conversion rate threshold, adjust the preset score threshold.

[0066] Specifically, when a user to be quantified is identified as a target user, they can be added to a preset target user list. This target user list can be used to indicate the user group that requires special attention (priority follow-up). For example, there can be multiple target user lists, and each target user list corresponds to a statistical cycle. That is, user objects identified as target users in the current statistical cycle are added to the same target user list, and in the next statistical cycle, they are added to another target user list.

[0067] It is understandable that for the user objects in the target user list, the telemarketing system will sell to these users according to the preset specific marketing strategy, and it will be considered a successful conversion when the user completes the purchase of a specific product.

[0068] Then, according to the preset statistical cycle, the conversion status of each user object in the target user list is counted to obtain the actual telemarketing conversion rate corresponding to the target user list (telemarketing conversion rate = number of successfully converted users / total number of users in the target user list).

[0069] Then, based on the comparison between the actual telesales conversion rate and the preset conversion rate threshold, the preset score threshold is adjusted. For example, assuming the average conversion rate of ordinary users is 30%, the conversion rate threshold can be set to a value higher than the average conversion rate, such as 50%. If the actual telesales conversion rate is greater than the preset first conversion rate threshold (e.g., 50%), that is, the conversion rate of users in the target user list is significantly higher than the average conversion rate of ordinary users, indicating that the quantitative strategy for target users is effective, the preset score threshold for screening target users can be appropriately lowered to a smaller value. This will allow more user objects to be identified as target users.

[0070] Accordingly, when the actual telesales conversion rate is less than the preset second conversion rate threshold (which can be set to a value smaller than the first conversion rate threshold and slightly higher than the average conversion rate, such as 35%), that is, the conversion rate of users in the target user list is not significantly improved compared to the average conversion rate of ordinary users, indicating that the quantitative strategy for target users is too risky. In this case, it is necessary to increase the preset score threshold for screening target users. In this way, truly high-potential target users can be screened through a more stringent strategy to improve the conversion rate of target users.

[0071] Based on this, the actual conversion rate of target users is counted periodically, and the score threshold for identifying target users is dynamically adjusted based on the comparison between the conversion rate and the preset threshold, thereby further improving the flexibility and accuracy of user quantification.

[0072] In some possible embodiments, step S501, determining a quantization score corresponding to the user to be quantized based on a comprehensive feature set using a preset user quantization model, may include: S5013: Determine a first quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model; S5014. Generate label features corresponding to the user to be quantified based on the telemarketing conversation text using a preset label generation algorithm; S5015. Concatenate the user portrait features and the label features to obtain a second comprehensive feature set; S5016: Determine a second quantization score corresponding to the user to be quantized based on the second comprehensive feature set using the user quantization model; S5017: Determine a quantization score corresponding to the user to be quantized based on the first quantization score and the second quantization score.

[0073] It should be noted that in addition to using the segmented fusion embedding features based on the conversation text as the basis for user quantification, a user quantification basis for the control group can also be further set.

[0074] Specifically, a preset tag generation algorithm is first used to directly generate tag features corresponding to the telemarketing conversation text. This is then combined with the user profile features and tag features to generate a second comprehensive feature set. The user quantification model is then used to determine the user's second quantization score based on this second comprehensive feature set. This second quantization score, along with the first quantization score determined by the user quantification model based on the comprehensive feature set, is then used to determine the final quantization score for the user being quantified.

[0075] Based on this, the quantization scores are determined by using segmented fusion embedding features based on the conversation text and label features based on the conversation text respectively, and the user's quantization results are determined comprehensively based on the two quantization scores, thereby further improving the accuracy of user quantization.

[0076] Please refer to Figure 2 , Figure 2 The block diagram of the user quantification device based on the embedded features of conversation text provided by some embodiments of the present application is shown. It should be understood that the user quantification device based on the embedded features of conversation text is similar to the above Figure 1 Corresponding to the method embodiment, the various steps involved in the above method embodiment can be executed. The specific functions of the user quantification device based on conversation text embedding features can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here.

[0077] Figure 2 The user quantification device based on the embedded features of conversational texts includes at least one software function module that can be stored in a memory in the form of software or firmware or fixed in the user quantification device based on the embedded features of conversational texts, and the user quantification device based on the embedded features of conversational texts includes: The data acquisition module 210 is used to obtain the user portrait features and telemarketing conversation text of the user to be quantified; Feature extraction module 220, for extracting multiple key text segments from the telemarketing conversation text based on a preset process node division rule, and obtaining embedded features for each key text segment; A feature aggregation module 230 is configured to perform weighted aggregation on each embedded feature based on preset weight information to obtain a conversation embedded feature; A feature concatenation module 240 is configured to concatenate user profile features and conversation embedding features to obtain a comprehensive feature set; The user quantization module 250 is configured to determine a quantization result corresponding to the user to be quantized based on a comprehensive feature set using a preset user quantization model.

[0078] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention. The user quantification device based on the embedded features of conversation text provided by the embodiment of the present invention can implement the user quantification method based on the embedded features of conversation text provided by any method embodiment of the present invention.

[0079] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method, and will not be described in detail here.

[0080] like Figure 3 As shown, some embodiments of the present application provide an electronic device 300, which includes: a memory 310, a processor 320, and a computer program stored in the memory 310 and executable on the processor 320, wherein the processor 320 reads the program from the memory 310 via a bus 330 and executes the program to implement a method of any embodiment of the user quantification method based on conversation text embedding features as described above.

[0081] Processor 320 can process digital signals and can include various computing architectures, such as a complex instruction set computer architecture, a reduced instruction set computer architecture, or an architecture that implements a combination of multiple instruction sets. In some examples, processor 320 can be a microprocessor.

[0082] The memory 310 can be used to store instructions executed by the processor 320 or data related to the execution of instructions. These instructions and / or data may include code for implementing some or all functions of one or more modules described in the embodiments of this application. The processor 320 of the embodiment of the present disclosure can be used to execute the instructions in the memory 310 to implement the method shown above. The memory 310 includes dynamic random access memory, static random access memory, flash memory, optical memory, or other memory known to those skilled in the art.

[0083] Some embodiments of the present application further provide a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method described in the method embodiment is executed.

[0084] Some embodiments of the present application further provide a computer program product, which, when running on a computer, enables the computer to execute the method described in the method embodiment.

[0085] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similarities between the various embodiments can be referred to in conjunction with each other. For device embodiments, since they are generally similar to method embodiments, their description is relatively simple, and for relevant details, reference can be made to the description of the method embodiments.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0087] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0088] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0089] The foregoing is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included within the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures.

[0090] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0091] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

Claims

1. A user quantification method based on conversation text embedding features, characterized in that: include: Obtain user portrait features and telemarketing conversation texts of the users to be quantified; Extracting multiple key text segments from the telemarketing conversation text based on a preset process node division rule, and obtaining embedding features of each key text segment; Performing weighted aggregation on each of the embedding features based on preset weight information to obtain a conversation embedding feature; Concatenating the user portrait features and the conversation embedding features to obtain a comprehensive feature set; A preset user quantization model is used to determine a quantization result corresponding to the user to be quantified based on the comprehensive feature set.

2. The user quantification method based on conversation text embedding features according to claim 1, characterized in that: The step of weighting and aggregating the embedded features based on preset weight information to obtain the conversation embedded features includes: Acquiring attribute information of each key text segment, and determining weight information corresponding to each key text segment according to the attribute information of each key text segment; The embedded features are weighted and aggregated based on the weight information to obtain a conversation embedded feature.

3. The user quantification method based on conversation text embedding features according to claim 2, characterized in that: The acquiring of the attribute information of each key text segment and determining the weight information corresponding to each key text segment according to the attribute information of each key text segment includes: Acquiring attribute information of each of the key text segments; wherein the attribute information includes at least one of a text length attribute, a dialogue clarity attribute, a dialogue efficiency attribute, and a user intent clarity attribute; The importance of each key text segment is determined based on the attribute information of each key text segment, and weight information corresponding to each key text segment is determined according to the importance of each key text segment.

4. The user quantification method based on conversation text embedding features according to claim 1, characterized in that: The determining of a quantization result corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model includes: Determining a quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model; When the quantization score is greater than a preset score threshold, the user to be quantified is identified as a target user.

5. The user quantification method based on conversation text embedding features according to claim 4 is characterized in that: The determining of a quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model includes: Determine a first target quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset first user quantization model, and determine a second target quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset second user quantization model; A quantization score corresponding to the to-be-quantized user is determined based on the first target quantization score and the second target quantization score.

6. The user quantification method based on conversation text embedding features according to claim 4, characterized in that: Also includes: Add the user object identified as the target user to the target user list; Counting the actual telesales conversion rate corresponding to the target user list according to a preset statistical period; Based on the comparison between the actual telesales conversion rate and the preset conversion rate threshold, the preset score threshold is adjusted.

7. The user quantification method based on conversation text embedding features according to claim 4, characterized in that: The determining of a quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model includes: Determining a first quantization score corresponding to the user to be quantized based on the comprehensive feature set using a preset user quantization model; Generate label features corresponding to the user to be quantified based on the telemarketing conversation text using a preset label generation algorithm; Concatenate the user portrait features and the tag features to obtain a second comprehensive feature set; Determining a second quantization score corresponding to the user to be quantized based on the second comprehensive feature set using the user quantization model; A quantization score corresponding to the to-be-quantized user is determined based on the first quantization score and the second quantization score.

8. An electronic device, characterized in that: The invention comprises 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 user quantification method based on conversation text embedding features according to any one of claims 1 to 7 can be implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the user quantification method based on conversation text embedding features according to any one of claims 1 to 7 is executed.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, it implements the user quantification method based on conversation text embedding features according to any one of claims 1 to 7.

Images