Business recommendation method and device, equipment, storage medium and product

Abstract

The invention discloses a business recommendation method and device, equipment, a storage medium and a product, and relates to the technical field of business marketing. According to the method and the device, by determining the first association feature, the fragment highly related to each service label can be extracted from the description information, so that the content of key modeling is determined, interference among different service labels is avoided, and the accuracy of a subsequently established user portrait is improved; the second association feature is obtained through the association relationship between any two service tags, implicit association which is not directly expressed in description information but actually exists can be inferred, and deeper internal features are explored, so that the accuracy of a subsequently established user portrait is further improved, and the user experience is improved. The technical problem of how to efficiently generate more accurate user portraits and realize more accurate service recommendation for users is solved.

Description

Technical Field

[0001] The present application relates to the field of business marketing technology, and in particular to a business recommendation method, device, equipment, storage medium and product. Background Art

[0002] In related technologies, in order to achieve more accurate business recommendations for users, it is necessary to generate corresponding portraits for users. For the task of generating user portraits in precision marketing scenarios, related technologies generally use traditional rule-based methods, which usually rely on manually designing multiple templates. This rule-based method usually summarizes the characteristics of user data, and then designs multiple template sets for data matching, thereby discovering the subject domain information contained in the user data.

[0003] However, for massive user data with multiple business tags, this method has poor scalability, requires a lot of manpower and time to maintain the template collection, and has low adaptability to the production environment.

[0004] Therefore, how to efficiently generate more accurate user portraits and make more precise business recommendations to users is a problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of this application is to provide a business recommendation method, device, equipment, storage medium and product, aiming to solve the technical problem of how to efficiently generate more accurate user portraits and make more precise business recommendations to users.

[0006] To achieve the above objectives, the present application proposes a service recommendation method, which includes: Get the target user's description information; Determining a first association feature between the description information and all the service tags in a service tag set; wherein the service tag set includes a plurality of predefined service tags; Determine an association relationship between any two of the service tags to obtain a second association feature of the service tag set; Based on the target feature, establishing a user profile of the target user; the target feature is obtained by fusing the first associated feature and the second associated feature; Based on the user portrait, a business recommendation result for the target user is obtained.

[0007] In some embodiments, determining the first association feature between the description information and all the service tags in the service tag set includes: The encoder of the BERT model is represented by a bidirectional encoder, and semantic encoding is performed on each of the service tags in the service tag set to obtain multiple first encoding vectors; The encoder of the BERT model is represented by a bidirectional encoder to perform semantic encoding processing on the description information to obtain a second encoding vector corresponding to each word in the description information; Calculating the similarity between each first encoding vector and each second encoding vector; All the similarities are merged with all the second encoding vectors to determine a first association feature between the description information and all the service tags in the service tag set.

[0008] In some embodiments, calculating the similarity between each first encoding vector and each second encoding vector comprises: By expression 1, calculate any first encoding vector With any second encoding vector Similarity ; The fusing all the similarities with all the second encoding vectors to determine the first association feature between the description information and all the service tags in the service tag set includes: By using expression 2, the similarity between any one of the first coding vectors and any one of the second coding vectors is merged with the second coding vector to obtain an intermediate correlation feature; Superimposing all the intermediate correlation features to obtain the first correlation feature; Among them, expression 1 is:

[0009] Expression 2 is: =

[0010] is any one of the first encoding vectors, m is the total number of the first encoding vectors, is any one of the second encoding vectors, n is the total number of the second encoding vectors, For the With the The similarity between them, y is the intermediate correlation feature.

[0011] In some embodiments, determining the association relationship between any two of the service tags to obtain the second association feature of the service tag set includes: For each of the service tags, based on the similarity corresponding to the service tags, determining a text segment corresponding to the service tag in the description information; wherein the window lengths of the text segments corresponding to each of the service tags are the same; Calculating the association weight between any two of the service tags based on the word segments in the text segment; Based on the association weight, a second association feature of the service tag set is determined.

[0012] In some embodiments, the calculating the association weight between any two of the service tags based on the word segments in the text segment includes: By using expression 3, the association weight between the first service label and the second service label is calculated, wherein the first service label is any one of the plurality of service labels, and the second service label is any one of the plurality of service labels except the first service label; Expression three is: =

[0013] in, is the association weight between the first service label i and the second service label j, k is the window length, Calculated by expression 4; Expression 4 is:

[0014] Among them, A is the preset multi-head self-attention matrix, is the hth word in the text segment corresponding to the first business label, is the hth word segment in the text segment corresponding to the second business label, and p is the number of heads in the preset multi-head self-attention matrix.

[0015] In some embodiments, determining the second association feature of the service tag set based on the association weight includes: Based on the association weight, calculating the intermediate association feature of the first service tag to the second service tag through expression five; Superimposing all the intermediate correlation features to obtain the second correlation feature; Expression five is: =

[0016] in, is the intermediate association feature, is the association weight between the first service label i and the second service label j, The first service tag is a plurality of word segments in the text segment corresponding to the first service tag.

[0017] In addition, to achieve the above purpose, the present application also proposes a service recommendation device, which includes: A data acquisition module is used to obtain description information of target users; An explicit modeling module, used to determine a first association feature between the description information and all the service tags in a service tag set; wherein the service tag set includes a plurality of predefined service tags; An implicit modeling module, used to determine the association relationship between any two of the service tags to obtain a second association feature of the service tag set; A portrait generation module, used to establish a user portrait of the target user based on a target feature; the target feature is obtained by fusing the first associated feature and the second associated feature; The business recommendation module is used to obtain the business recommendation result of the target user based on the user portrait.

[0018] In addition, to achieve the above objectives, the present application also proposes a business recommendation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the business recommendation method described above.

[0019] In addition, to achieve the above objectives, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the business recommendation method described above are implemented.

[0020] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the service recommendation method described above are implemented.

[0021] One or more technical solutions proposed in this application have at least the following technical effects: By determining the first association feature, fragments that are highly relevant to each business tag can be extracted from the description information, thereby determining the content for key modeling, avoiding interference between different business tags, and improving the accuracy of the user portrait subsequently established; by obtaining the second association feature through the association relationship between any two of the business tags, it is possible to infer implicit associations that are not directly expressed in the description information but actually exist, and discover deeper intrinsic features, thereby further improving the accuracy of the user portrait subsequently established and achieving business recommendations that are more in line with the actual needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 A schematic diagram showing a process flow of a service recommendation method provided by an embodiment of the present application is shown; Figure 2 A schematic diagram showing the structure of a business recommendation device provided in an embodiment of the present application is shown; Figure 3 A schematic diagram of the structure of a business recommendation device provided in an embodiment of the present application is shown.

[0025] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0026] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0027] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0028] The main solution of the embodiment of the present application is: obtain the description information of the target user; determine the first association feature between the description information and all the services in the service set; wherein the service set includes multiple pre-defined services; determine the association relationship between any two services to obtain the second association feature of the service set; based on the target feature, establish a user profile of the target user; the target feature is obtained by fusing the first association feature and the second association feature; based on the user portrait, obtain the service recommendation result for the target user.

[0029] In the related art, the generation of user portraits can be roughly divided into three categories. Among them, the first category belongs to the traditional rule-based method, which usually relies on manually designing multiple templates, or combining machine learning to design kernel functions (used to map data from the original feature space to a higher-dimensional feature space in order to find the decision boundary in this high-dimensional space). This rule-based method usually summarizes the characteristics of user data, and then designs multiple template sets for data matching, and then discovers the subject domain information contained in the user data. However, for massive, multi-business subject domain scale user data, this method has poor scalability, and requires a lot of manpower and time to maintain the template set, and has low adaptability to the production environment; the second category belongs to the vector-based method, which usually decomposes and converts user data into word vectors, and analyzes based on word vectors, but this method only preliminarily completes the mapping from word to vector, and does not really model the characteristics of the word; the third category is a method based on deep learning, which usually combines neural networks to automatically analyze and extract features of user data, which can effectively alleviate the defects of traditional methods, but currently lacks the association between different subject domain features in user data. Mining.

[0030] In summary, how to efficiently generate more accurate user portraits and make more precise business recommendations to users is a problem that needs to be solved urgently.

[0031] Based on this, the present application provides a solution that can more efficiently generate more accurate user portraits, taking into account the correlation between businesses, thereby achieving more precise business marketing.

[0032] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or a business recommendation device capable of realizing the above functions, etc. The following takes the business recommendation device as an example to illustrate this embodiment and the following embodiments.

[0033] Reference Figure 1 , Figure 1 The flowchart of the service recommendation method provided by an embodiment of the present application is shown. The service recommendation method can be applied to a service recommendation device, and includes the following steps S110 to S150: Step S110, obtaining description information of the target user.

[0034] The target user refers to the user to whom the business marketing is to be conducted. It is understandable that the attribute information of the user generally includes multiple dimensions, such as the user's name, age, gender, nature of work, income range, permanent location, marital relationship, etc. In some embodiments, the attribute information of the target user may be pre-stored in a database, and when needed, the attribute information of the target user may be directly retrieved from the database by using the user's name and other information. In other embodiments, the attribute information of the target user may also be input by the staff in real time, which is not limited in this embodiment.

[0035] In some implementations, the attribute information of the target user may be concatenated in natural language using a large language model, thereby generating a long text containing semantic information and being logically coherent, which may be used as the description information of the target user.

[0036] For example, in one example, the description information is "the user is a high-value user in XX city, lives in the urban area, is an active user of voice calls, and has called the customer service hotline in the past three months."

[0037] Step S120: determining a first correlation feature between the description information and all the service tags in the service tag set.

[0038] The business tag set includes multiple predefined business tags, and the content and quantity of the business tags included in the business tag set are pre-set by the staff. Business tags refer to the subject dimension related to a specific business. Each business has at least one business tag (for example, the customer service business has the business tag "customer service"), and different businesses can have overlapping business tags.

[0039] The user's description information may include a part corresponding to a service tag. For example, a service tag may be "customer service". If the description information is "the user is a high-value user in XX city, lives in the urban area, is an active user of voice calls, and has called the customer service hotline in the past three months", the part corresponding to the service tag "customer service" is "has called the customer service hotline in the past three months".

[0040] It is understandable that different services have different focuses on the description information of the user, that is, the description information of the target user has different focuses under different service tags.

[0041] In the above example, assuming that the description information is "The user is a high-value user in XX city, lives in the urban area, is an active user of voice calls, and has called the customer service hotline in the past three months", for the service tag "User Usage Behavior", "Is an active user of voice calls" is the key focus. For the service tag "Customer Service", "Did call the customer service hotline in the past three months" is the key focus.

[0042] In the related art, for different services, the features obtained from the description information of users are the same, which makes it impossible to generate user portraits in a targeted manner and it is difficult to ensure the accuracy of service recommendations. Therefore, in order to further improve the accuracy of subsequent user portrait generation, this embodiment designs to explore the duality between the description information context and different service labels, and use service labels as guidance signals to enable the model to learn better text representation.

[0043] In some implementations, the description information may be semantically encoded by an encoder of a BERT model represented by a bidirectional encoder to obtain a second encoding vector corresponding to each word in the description information.

[0044] Specifically, by embedding the description information through the bidirectional encoder representation BERT model (hereinafter referred to as the BERT model), the contextual relationship in the description information can be further explored.

[0045] First, to meet the input requirements of the BERT model, the long text can be segmented. Secondly, the word segmentation result is input into the embedding layer. The embedding layer inserts two special tags into it, converting it into "[CLS] User is a high-value user in XX City, lives in the urban area, is an active user of voice calls, and has called the customer service hotline in the past three months. [SEP]". Among them, "[CLS]" represents the beginning of the text and "[SEP]" represents the end of the text.

[0046] Then, each word in the text is mapped to a low-dimensional vector based on the original BERT dictionary. The embedding function includes three parts: word embedding, position embedding, and segment embedding. "[CLS] user is a high-value user in XX city, lives in the urban area, is an active user of voice calls, and has called the customer service hotline in the past three months. [SEP]" is abstracted as: d={ , , …, }, where n is the total number of participles (including inserted special tags). For each participle , you can use the participle Mapped to a word embedding vector , the corresponding position Mapped to a position embedding vector ; Participle The embedding vector corresponding to the segment is , then for , which can be transformed into:

[0047] The embedding representation of the entire text sequence is:

[0048] Furthermore, based on the encoding process of the embedding matrix, BERT's encoding module is implemented using the self-attention method in the Transformer architecture, and combined with the "multi-head" strategy to simultaneously capture features of different dimensions. By processing the word embedding matrix representation through the multi-head self-attention mechanism, the long-range semantic association of the text can be captured, and the user data features can be fully modeled. The calculation formula of the single-head self-attention mechanism is as follows:

[0049] in, Preset weight matrices for queries, keys, and values ​​respectively.

[0050]

[0051] in, is the dimension of the key matrix.

[0052]

[0053] () is the activation function, is the preset bias vector, is the preset weight matrix of the feedforward neural network. The multi-head self-attention mechanism is obtained by concatenating the output h of each single head.

[0054] Finally, connect a feed-forward neural network layer to get the final encoded representation:

[0055] H is the second encoding vector set, which contains the second encoding vector corresponding to each word segment.

[0056] It is understandable that the use of the BERT model to obtain the first encoding vector does not have too many restrictions on the content of the first encoding vector compared to the template, rule or vector-based methods in the related art, and can be migrated between different business subject domains, with stronger versatility. In addition, the BERT model models the semantic features of each word in the description information in the current context through context analysis of the multi-head self-attention mechanism, effectively avoiding semantic ambiguity problems, and can also capture long-distance dependency information in the description information, laying a good foundation for subsequent steps.

[0057] In some implementations, similar to the aforementioned processing method, the encoder of the BERT model can be represented by a bidirectional encoder to perform semantic encoding processing on each business tag in the business tag set to obtain multiple first encoding vectors.

[0058] After using the encoding layer in the BERT model to learn the features of the business label, the corresponding encoding representation can be obtained:

[0059] in, is the first encoding vector corresponding to the j-th service label in the service label set.

[0060] In some embodiments, since the service tag is often triggered by a similar context in the text, there is a direct correlation between the service tag and the similar fragment in the text. In some embodiments, the similarity between each first encoding vector and each second encoding vector can be calculated, and all similarities can be merged with all second encoding vectors to determine the first correlation feature between the description information and all services in the service set.

[0061] Specifically, the similarity between any first coding vector and any second coding vector can be calculated by expression 1; all similarities can be merged with all second coding vectors by expression 2 to obtain the intermediate correlation feature; All intermediate correlation features are superimposed to obtain the first correlation feature; Among them, expression 1 is:

[0062] Expression 2 is: =

[0063] is the first encoding vector, m is the total number of first encoding vectors, is the second encoding vector, n is the total number of second encoding vectors, for and The similarity between them, y is the intermediate correlation feature.

[0064] Through expression 1 and expression 2, the correlation between any business tag and any word segment (text fragment) can be calculated. Based on any business tag and any word segment, the text representation of the target user's description information under different business tags can be obtained.

[0065] In this embodiment, through the above step S120, important fragments of the description information for different business tags are analyzed, and specific text representations are generated for different business tags. Compared with the single text representation in the related art, this embodiment can more accurately capture the content of concern under different business tags, avoid interference of other business tags on the current business tag, and improve the accuracy of subsequent user portrait generation.

[0066] Step S130: determine the association relationship between any two service tags, and obtain a second association feature of the service tag set.

[0067] Through step S120, the business tag is used as a guide to effectively mine the relevant text fragments of each business tag. In this process, the feature modeling of each business tag is carried out independently, and there is no interaction between different business tags. However, in actual marketing scenarios, the information of a certain business tag is rarely considered separately. Instead, the user is taken as the main factor to integrate the business subject domains of multiple dimensions to generate a user integration portrait. By integrating different business subject domains, it is possible to infer implicit associations that are not directly expressed in the description information but actually exist, thereby improving the accuracy of marketing. For example, by combining the user's descriptive information in the two business subject domains of "usage behavior" and "service touchpoints", the user's activity level and service expectations can be inferred, thereby improving marketing strategies and increasing user satisfaction. Therefore, exploring the interaction between different business tags can capture potential semantic association features and further improve the accuracy of user portraits.

[0068] Based on this, this embodiment is designed to determine the association relationship between any two service tags through step S130, so as to obtain the second association feature of the entire service tag set.

[0069] Specifically, for each service tag, a text segment corresponding to the service tag may be determined in the description information based on the similarity corresponding to the service tag.

[0070] By similarity By comparing the description information, we can know which content is the most important for the selected business tag. After determining the location of the important content, we can take the starting point of the important content in the description text as the starting point and cut it according to the set window length to obtain the text segment corresponding to each business tag, and these text segments all have the same window length.

[0071] For any two service labels , , and the corresponding text fragments are represented as follows:

[0072]

[0073] Where k is the window length.

[0074] In some implementations, after obtaining the text segment, the association weight between any two business tags can be calculated based on the word segments in the text segment. Through the multi-head self-attention mechanism, the description information in each business tag is learned to pay attention to the dependency of other business tags, thereby obtaining the word segmentation level weight distribution between business tag pairs.

[0075] Specifically, the association weight between the first service label and the second service label may be calculated by expression three, wherein the first service label is any one of the multiple service labels, and the second service label is any one of the multiple service labels except the first service label; Expression three is: =

[0076] in, is the association weight between the first service label i and the second service label j, Calculated by expression 4; Expression 4 is:

[0077] Among them, A is the preset multi-head self-attention matrix, is the hth word in the text segment corresponding to the first business label, is the hth word segment in the text segment corresponding to the second business label, and p is the number of heads in the preset multi-head self-attention matrix.

[0078] After obtaining the association weight of the first service tag with each of the other service tags, the intermediate association feature corresponding to the first service tag may be determined based on the corresponding association weights.

[0079] Specifically, the intermediate correlation feature of the first service label to the second service label can be calculated by expression 5; all the intermediate correlation features are superimposed to obtain the second correlation feature corresponding to the entire service label set; Expression five is: =

[0080] in, is the intermediate association feature, is the association weight between the first service label i and the second service label j, The first service tag is a plurality of word segments in the text segment corresponding to the first service tag.

[0081] Step S140: Create a user profile of the target user based on the target features.

[0082] In some implementations, the first associated feature and the second associated feature may be fused together, and the overall feature summary of the fused text may be added to obtain the target feature. For example, the target feature may be characterized as: =

[0083] in, is the first associated feature, is the second associated feature, The special symbol [CLS] added by the BERT encoding layer during the embedding process represents the overall feature summary of the fused text, thereby ensuring the integrity of the fused text (i.e., the target feature).

[0084] In some implementations, after obtaining the target features corresponding to the target user, the target features may be input into a model to generate a user profile of the target user.

[0085] Generating a user profile can include the following steps: First, based on the target features, the probability of each business tag in the business tag set is calculated:

[0086] =

[0087] in, Indicates the first Business tags, is the target feature, is the probability layer parameter, Represents an activation function, such as tanh().

[0088] Subsequently, the categorical cross-entropy loss function is used to train and optimize the model:

[0089] in, is the number of business tags in the business tag set; =0 or 1, used to indicate whether the text contains the Business tags; The model predicts that there is The probability of a business label.

[0090] As a feasible example, the model can use the AdamW optimizer to optimize the hyperparameters of the model. This embodiment uses BERT as the encoder. During the training process, the classification accuracy of the model is improved by minimizing the loss function. The model parameters can be fine-tuned according to the performance of the model during the gradient update process. All hyperparameter tuning is completed on the data validation set. Specifically, the learning rate of the model can be selected from {1e-5; 2e5; 5e-5; 1e-4}; the linear warm-up can be set to the first 6% steps, and the linear decay can be set to 0; the training cycle of the model can be set to 20 rounds, and the early termination strategy can be set. The basis for termination can be the accuracy on the data validation set.

[0091] In this embodiment, the process of "generating user portraits" is abstracted into a multi-label classification task, and the model is trained and fine-tuned using supervised learning. During the model debugging process, the model is trained using annotated actual production data, which makes up for the lack of professional knowledge in general large language models and can effectively enhance the professional capabilities of the model.

[0092] Step S150, obtaining a business recommendation result for the target user based on the user portrait.

[0093] The business recommendation results may include marketing methods, marketing content, etc., which may be different in different application scenarios and are not limited in this embodiment. Through the user portrait obtained in step S140, more detailed feature information contained in the description information of the target user can be known, so as to accurately recommend business content that better meets the user's needs.

[0094] Specifically, this embodiment can be mainly applied to precision marketing services. By analyzing the usage behavior, contact behavior, etc. of the existing user description information, the business subject domain in which the user is mainly involved can be determined, so as to provide users with accurate packages, after-sales, activities and other solutions to improve customer satisfaction.

[0095] This embodiment provides a service recommendation method. By determining the first association feature, segments highly related to each service tag can be extracted from the description information, thereby determining the content to be focused on modeling, avoiding interference between different service tags, and improving the accuracy of the user portrait subsequently established. By obtaining the second association feature through the association relationship between any two of the service tags, it is possible to infer implicit associations that are not directly expressed in the description information but actually exist, and explore deeper intrinsic features, thereby further improving the accuracy of the user portrait subsequently established and achieving service recommendations that are more in line with the actual needs of users.

[0096] This application also provides a business recommendation device, please refer to Figure 2 , the business recommendation device 100 includes: The data acquisition module 110 is used to acquire the description information of the target user; An explicit modeling module 120, configured to determine a first correlation feature between the description information and all the service tags in a service tag set; wherein the service tag set includes a plurality of predefined service tags; An implicit modeling module 130, configured to determine an association relationship between any two of the service tags to obtain a second association feature of the service tag set; A portrait generation module 140 is used to establish a user portrait of the target user based on a target feature; the target feature is obtained by fusing the first associated feature and the second associated feature; The service recommendation module 150 is used to obtain the service recommendation result of the target user based on the user portrait. The service recommendation device 100 provided in the present application adopts the service recommendation method in the above embodiment, which can solve the technical problem of how to efficiently generate a more accurate user portrait and implement more accurate service recommendations for users. Compared with the prior art, the beneficial effects of the service recommendation device 100 provided in the present application are the same as the beneficial effects of the service recommendation method provided in the above embodiment, and the other technical features in the service recommendation device 100 are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0097] The present application provides a business recommendation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the business recommendation method in the above-mentioned embodiment one.

[0098] Reference below Figure 3 , which shows a schematic diagram of the structure of a service recommendation device suitable for implementing the embodiment of the present application. The service recommendation device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The business recommendation device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0099] like Figure 3 As shown, the business recommendation device 200 may include a processing device 210 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 220 or a program loaded from a storage device 230 to a random access memory (RAM: Random Access Memory) 240. In RAM240, various programs and data required for the operation of the business recommendation device are also stored. The processing device 210, ROM220 and RAM240 are connected to each other through a bus 250. An input / output (I / O) interface 260 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 260: an input device 270 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 280 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 230 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 290. The communication device 290 can allow the business recommendation device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a business recommendation device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided instead.

[0100] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from the storage device 230, or installed from the ROM 220. When the computer program is executed by the processing device 210, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0101] The service recommendation device provided by the present application adopts the service recommendation method in the above embodiment, which can solve the technical problem of how to efficiently generate a more accurate user portrait and implement more accurate service recommendations for users. Compared with the prior art, the beneficial effects of the service recommendation device provided by the present application are the same as the beneficial effects of the service recommendation method provided by the above embodiment, and the other technical features in the service recommendation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0102] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0103] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0104] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the service recommendation method in the above-mentioned embodiment.

[0105] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0106] The computer-readable storage medium may be included in the service recommendation device; or may exist independently without being installed in the service recommendation device.

[0107] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the business recommendation device, the business recommendation device can write computer program codes for performing the operations of the present application in one or more programming languages ​​or a combination thereof. The above-mentioned programming languages ​​include object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0108] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession 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 square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0109] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0110] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned service recommendation method, and can solve the technical problem of how to efficiently generate more accurate user portraits and achieve more accurate service recommendations for users. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the service recommendation method provided in the above-mentioned embodiment, and will not be repeated here.

[0111] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned service recommendation method when executed by a processor.

[0112] The computer program product provided by this application can solve the technical problem of how to efficiently generate more accurate user portraits and implement more accurate business recommendations for users. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the business recommendation method provided by the above embodiment, and will not be repeated here.

[0113] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A business recommendation method, characterized in that: The service recommendation method comprises: Get the target user's description information; Determining a first association feature between the description information and all the service tags in a service tag set; wherein the service tag set includes a plurality of predefined service tags; Determine an association relationship between any two of the service tags to obtain a second association feature of the service tag set; Based on the target feature, establishing a user profile of the target user; the target feature is obtained by fusing the first associated feature and the second associated feature; Based on the user portrait, a business recommendation result for the target user is obtained.

2. The service recommendation method according to claim 1, characterized in that: The determining of the first association feature between the description information and all the service tags in the service set includes: The encoder of the BERT model is represented by a bidirectional encoder, and semantic encoding is performed on each of the service tags in the service tag set to obtain multiple first encoding vectors; The encoder of the BERT model is represented by a bidirectional encoder to perform semantic encoding processing on the description information to obtain a second encoding vector corresponding to each word in the description information; Calculating the similarity between each first encoding vector and each second encoding vector; All the similarities are merged with all the second encoding vectors to determine a first association feature between the description information and all the service tags in the service tag set.

3. The service recommendation method according to claim 2, characterized in that: The calculating the similarity between each first encoding vector and each second encoding vector comprises: By using expression 1, the similarity between any first encoding vector and any second encoding vector is calculated; The fusing all the similarities with all the second encoding vectors to determine the first association feature between the description information and all the service tags in the service tag set includes: By using expression 2, all the similarities are merged with all the second encoding vectors to obtain an intermediate correlation feature; Superimposing all the intermediate correlation features to obtain the first correlation feature; Among them, expression 1 is: Expression 2 is: = is any one of the first encoding vectors, m is the total number of the first encoding vectors, is any one of the second encoding vectors, n is the total number of the second encoding vectors, For the With the The similarity between them, y is the intermediate correlation feature.

4. The service recommendation method according to claim 3, characterized in that: The determining the association relationship between any two of the service tags to obtain a second association feature of the service tag set includes: For each of the service tags, based on the similarity corresponding to the service tags, determining a text segment corresponding to the service tag in the description information; wherein the window lengths of the text segments corresponding to each of the service tags are the same; Calculating the association weight between any two of the service tags based on the word segments in the text segment; Based on the association weight, a second association feature of the service tag set is determined.

5. The service recommendation method according to claim 4, characterized in that: The calculating the association weight between any two of the service tags based on the word segments in the text segment includes: By using expression 3, the association weight between the first service label and the second service label is calculated, wherein the first service label is any one of the plurality of service labels, and the second service label is any one of the plurality of service labels except the first service label; Expression three is: = in, is the association weight between the first service label i and the second service label j, k is the window length, Calculated by expression 4; Expression 4 is: Among them, A is the preset multi-head self-attention matrix, is the hth word in the text segment corresponding to the first business label, is the hth word segment in the text segment corresponding to the second business label, and p is the number of heads in the preset multi-head self-attention matrix.

6. The service recommendation method according to claim 5, characterized in that: The determining, based on the association weight, a second association feature of the service set includes: Based on the association weight, calculating the intermediate association feature of the first service tag to the second service tag through expression five; Superimposing all the intermediate correlation features to obtain the second correlation feature; Expression five is: = in, is the intermediate association feature, is the association weight between the first service label i and the second service label j, The first service tag is a plurality of word segments in the text segment corresponding to the first service tag.

7. A business recommendation device, characterized in that: The business recommendation device comprises: A data acquisition module is used to obtain description information of target users; An explicit modeling module, used to determine a first association feature between the description information and all the service tags in a service tag set; wherein the service tag set includes a plurality of predefined service tags; An implicit modeling module, used to determine the association relationship between any two of the service tags to obtain a second association feature of the service tag set; A portrait generation module, used to establish a user portrait of the target user based on a target feature; the target feature is obtained by fusing the first associated feature and the second associated feature; The business recommendation module is used to obtain the business recommendation result of the target user based on the user portrait.

8. A business recommendation device, characterized in that: The service recommendation device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the service recommendation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the service recommendation method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the service recommendation method according to any one of claims 1 to 6 are implemented.

Images