Intelligent product recommendation method and device based on social data, equipment and medium

Abstract

The application provides a social data-based intelligent product recommendation method and device, an electronic device and a computer readable storage medium. The method comprises the following steps: constructing a plurality of scene lines for identifying user demand, wherein the scene lines comprise a plurality of story nodes, each story node comprises a story label and a preset semantic classification model; obtaining social text data of a user, determining the matching degree of the social text data and each scene line according to the semantic classification model of the plurality of scene lines, and taking the scene line meeting the matching requirement as a first story line; determining a first word vector of each word segmentation in the social text data, determining a second word vector of a preset keyword of each story label in each first story line, and determining the similarity between the social text data and the first story line; taking the first story line meeting a preset threshold requirement as a target story line to determine a target recommended product. The application can mine the real demand of the user and improve the accuracy of product recommendation.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation technology, and in particular to an intelligent product recommendation method, device, electronic device, and computer-readable storage medium based on social data. Background Technology

[0002] In current product recommendation applications based on financial transactions, intelligent recommendation is a crucial module. In practical applications, intelligent financial product recommendation systems not only need to uncover users' financial needs from historical financial transaction data, but also need to fully explore user characteristics from various data sources to ensure that recommended financial products meet the potential users' needs to the greatest extent possible.

[0003] Traditional financial product recommendation systems mostly rely on users' historical financial transaction records, such as investment records, insurance policy records, and transaction breakpoints. The advantage of using historical transaction records for user analysis is that the data source is relatively direct, allowing for intuitive analysis of the scope of a user's transactions, key areas where the user lacks protection, and enabling targeted marketing based on the analysis results. The main problem with these traditional financial product recommendation systems is their limited user data source. They cannot obtain a true picture of users' lives from historical transaction data, making it difficult to uncover users' real needs for financial products in their daily lives, resulting in low accuracy in financial product recommendations. Summary of the Invention

[0004] In view of the above, it is necessary to propose a method, device, electronic device and computer-readable storage medium for intelligent product recommendation based on social data, so as to solve the technical problem of low accuracy in financial product recommendations.

[0005] In a first aspect, embodiments of this application provide an intelligent product recommendation method based on social data, the method comprising:

[0006] Construct multiple scene lines for identifying user needs. Each scene line includes several story nodes, each story node includes a story tag and a preset semantic classification model, and each scene line corresponds to at least one recommended product.

[0007] Obtain the user's social text data, determine the matching degree between the social text data and each scene line based on the semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line;

[0008] Determine the first lexical vector for each word segment in the social text data, and determine the second lexical vector for each preset keyword of each story tag in each first storyline;

[0009] The similarity between the social text data and the first storyline is determined based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline.

[0010] The first storyline whose similarity meets the preset threshold requirement is taken as the target storyline, and the target recommended product is determined based on the target storyline.

[0011] In some embodiments, the semantic classification model includes a binary classification model; determining the matching degree between the social text data and each scene line based on the semantic classification model of the plurality of scene lines includes:

[0012] The social text data is input into each of the binary classification models of the scene line for classification, and the matching probability output by each binary classification model is obtained and used as the matching probability of the corresponding story node.

[0013] The matching probabilities of all story nodes in the scene line are combined and distributed to obtain the matching probability between the scene line and the social text data, which is then used as the matching degree.

[0014] In some embodiments, the binary classification model includes a BERT model and an ensemble classifier connected to the backend of the BERT model, wherein the matching probability of the story node is the average of the classification probabilities output by all sub-classifiers in the ensemble classifier.

[0015] In some embodiments, the training process of the BERT model includes:

[0016] Obtain a social text training dataset, wherein the samples in the social text training dataset are labeled with positive / negative sample categories;

[0017] The social text training dataset is input into the BERT model for encoding, and the predicted classification result is output through the backend ensemble classifier. The predicted classification result is compared with the positive / negative sample category of the corresponding sample to determine the prediction error.

[0018] The model parameters of the BERT model are adjusted according to the prediction error, and gradient descent learning is performed on the encoding part of the BERT model. Ensemble learning is performed on the back-end ensemble classifier. The training is iterated until the prediction error is less than a preset error threshold, and a fully trained BERT model is obtained.

[0019] In some embodiments, each story tag has at least one preset keyword; determining the first lexical vector of each word segment in the social text data, and determining the second lexical vector of the preset keyword of each story tag in each first storyline, includes:

[0020] The social text data is segmented into words, and each segmented word is converted into a corresponding first word vector using the preset word vector model Word2vec.

[0021] Using the pre-defined word vector model Word2vec, all pre-defined keywords of each story tag in the first storyline are converted into corresponding second word vectors.

[0022] In some embodiments, determining the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline includes:

[0023] Using the cosine similarity metric, the shortest distance from each first word vector of the social text data to all second word vectors of the first storyline is determined, and the average of the shortest distances corresponding to all first word vectors is taken as the similarity between the social text data and the first storyline.

[0024] In some embodiments, the recommended products include at least one service product in at least one financial theme.

[0025] Secondly, embodiments of this application provide an intelligent product recommendation device based on social data, the device comprising:

[0026] Construct scene units to build multiple scene lines for identifying user needs. Each scene line includes several story nodes, each story node includes a story tag and a preset semantic classification model, and each scene line corresponds to at least one recommended product.

[0027] The matching data unit is used to acquire the user's social text data, determine the matching degree between the social text data and each scene line according to the semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line.

[0028] A vector unit is defined to determine the first lexical vector of each word segment in the social text data, and to determine the second lexical vector of the preset keyword of each story tag in each first storyline;

[0029] The similarity calculation unit is used to determine the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline.

[0030] The product recommendation unit is used to select the first storyline whose similarity meets the preset threshold requirement as the target storyline, and to determine the target recommended product based on the target storyline.

[0031] Thirdly, embodiments of this application also provide an electronic device, the electronic device comprising:

[0032] Memory, storing at least one instruction;

[0033] The processor executes instructions stored in the memory to implement the intelligent product recommendation method based on social data.

[0034] This application also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the aforementioned intelligent product recommendation method based on social data.

[0035] In summary, this application constructs multiple scenario lines that target user needs, then semantically matches the user's social text data with each story node of each scenario line. This allows for the selection of several relatively matching first story lines. The social text data is then segmented, and each segmented word and the story tag keywords of each first story line are precisely filtered using word vector metrics to obtain the target story line and determine the target recommended product. This effectively improves the accuracy of product recommendations and helps enhance the marketing efficiency of subsequent marketing activities. Attached Figure Description

[0036] Figure 1 This is a flowchart of a preferred embodiment of the intelligent product recommendation method based on social data involved in this application.

[0037] Figure 2 This is a functional block diagram of a preferred embodiment of the intelligent product recommendation device based on social data involved in this application.

[0038] Figure 3 This is a schematic diagram of the structure of an electronic device that is a preferred embodiment of the intelligent product recommendation method based on social data involved in this application. Detailed Implementation

[0039] To better understand the purpose, features, and advantages of this application, a detailed description of the application is provided below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of this application can be combined with each other. Numerous specific details are set forth in the following description to provide a thorough understanding of this application; the described embodiments are only a part of the embodiments of this application, and not all of them.

[0040] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0042] This application provides a method for recommending intelligent products based on social data, which can be applied to one or more electronic devices. An electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0043] Electronic devices can be any electronic product that allows human-computer interaction, such as personal computers, tablets, smartphones, personal digital assistants (PDAs), game consoles, interactive network television (IPTV), smart wearable devices, etc.

[0044] Electronic devices may also include network devices and / or user devices. The network devices include, but are not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0045] The networks in which electronic devices are located include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, and virtual private networks (VPNs).

[0046] like Figure 1The diagram shown is a flowchart of a preferred embodiment of the intelligent product recommendation method based on social data according to this application. The order of the steps in this flowchart can be changed, and some steps can be omitted, depending on different needs.

[0047] Step S10: Construct multiple scene lines for identifying user needs. Each scene line includes several story nodes. Each story node includes a story tag and a preset semantic classification model. Each scene line corresponds to at least one recommended product.

[0048] It is understandable that all story nodes across all scene lines can be shared. A scene line can be a Markov chain composed of a series of pre-defined semantic classification models. A single Markov chain can accurately locate a scene with user demand, while different Markov chains, i.e., scene lines, locate different user demands. In addition, each story node carries a story tag with semantic information, and each story tag can be expanded with multiple pre-defined keywords to represent it. Thus, the pre-defined keywords of the story tag of each story node can form an independent keyword library for subsequent accurate matching.

[0049] In an optional embodiment, the recommended products include at least one service product in at least one financial theme.

[0050] The financial theme can include insurance, wealth management, lending, health, and investment. A financial theme can include multiple categories of service products. For example, the insurance theme includes health insurance, car insurance, and pet insurance.

[0051] In this optional embodiment, a scene line can locate a recommended product, or a recommended product can be located by multiple scene lines individually.

[0052] For example, consider the scenario line for pet insurance as a service product within the insurance theme: Scenario Line [Friends have pets]: Story Node 1 [Friends], Story Node 2 [Raising], Story Node 3 [Pets]. Here, "Friends have pets" is a scenario that identifies users with potential pet insurance needs. "Friends," "Raising," and "Pets" can be keywords in different story tags. "Friends" can also be keywords representing personal information such as relatives, names, colleagues, and classmates. "Raising" can also be keywords representing semantic information related to pet interaction, such as raising, feeding, adopting, playing with, taking care of, and walking. "Pets" can also be keywords representing semantic information about pets, such as cats, dogs, Tibetan mastiffs, and specific pet names.

[0053] In this way, by constructing multiple scene lines consisting of a series of story nodes, we can uncover users' real needs and effectively improve the accuracy of social data product recommendations.

[0054] Step S11: Obtain the user's social text data, determine the matching degree between the social text data and each scene line according to the preset semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line.

[0055] It is understood that the social text data can be text data directly entered by the user, or text data converted from voice data entered by the user.

[0056] In an optional embodiment, the preset semantic classification model includes a binary classification model; determining the matching degree between the social text data and each scene line based on the preset semantic classification model of the multiple scene lines includes:

[0057] The social text data is input into each of the binary classification models of the scene line for classification, and the matching probability output by each binary classification model is obtained and used as the matching probability of the corresponding story node.

[0058] The matching probabilities of all story nodes in the scene line are combined and distributed to obtain the matching probability between the scene line and the social text data, which is then used as the matching degree.

[0059] In this optional embodiment, social text data is input into each binary classification model of each scene line. Based on the story label of each story node, the binary classification model outputs the semantic matching probability between the social text data and the corresponding story label. Then, the matching probabilities of all story nodes in the corresponding scene line are jointly distributed to obtain the matching probability of the scene line. The specific calculation formula is as follows:

[0060] P(sl,text)=P(sp1|text)*P(sp2|text)*…*P(sp n |text)*P(text);

[0061] Among them, sp n Let be the nth story node, `text` be the social text data, `sl` be the scene line, and `P(text)` be the prior probability of the binary classification model identifying and classifying the social text data. The matching probability of each story node is P(sp). n |text), P(sl,text) is the matching probability of a scene line, which is the product of the matching probabilities of all story nodes on that scene line.

[0062] Furthermore, the matching probability of all scene lines with social text data is used as the matching degree, and all scene lines are sorted from highest to lowest matching degree. The scene lines with the highest matching degree are selected as the first story line. The preset number of scene lines can be set according to actual conditions and is not limited here.

[0063] In this way, a primary storyline that resonates with the user can be identified, facilitating precise targeting in the next step.

[0064] In an optional embodiment, the binary classification model includes a BERT model and an ensemble classifier connected to the backend of the BERT model, wherein the matching probability of the story node is the average of the classification probabilities output by all sub-classifiers in the ensemble classifier.

[0065] The Bert (Bidirectional Encoder Representations from Transformer) model is a pre-trained language representation model that learns a deep fusion feature vector from the input text sequence and then applies the learned deep fusion feature vector to the classification task of the ensemble classifier.

[0066] In this optional embodiment, given the social text data, the matching probability P(sp) of each story node is... n (|text) represents the classification probability output by an independent BERT model. The BERT model is currently a superior pre-trained model. In this embodiment, a BERT model trained on an open Chinese corpus is used as the basis for pre-training and fine-tuning on a social text training dataset. It should be noted that because the existing social text training dataset has an extremely uneven distribution of positive and negative samples, with the number of negative samples far exceeding the number of positive samples, this embodiment integrates an ensemble classifier at the backend of the BERT model. Based on the number of sub-classifiers, each sub-classifier is trained using a method of evenly distributing negative samples and fully sampling positive samples. The BERT model's encoding part is trained using gradient descent. Finally, the classification probability output by a single BERT model is the average of the classification probabilities output by all sub-classifiers. This effectively solves the problem of imbalanced positive and negative samples and helps improve the classification accuracy of the BERT model.

[0067] In this optional embodiment, the training process of the BERT model includes:

[0068] Obtain a social text training dataset, wherein the samples in the social text training dataset are labeled with positive / negative sample categories;

[0069] The social text training dataset is input into the BERT model for encoding, and the predicted classification result is output through the backend ensemble classifier. The predicted classification result is compared with the positive / negative sample category of the corresponding sample to determine the prediction error.

[0070] The model parameters of the BERT model are adjusted according to the prediction error, and gradient descent learning is performed on the encoding part of the BERT model. Ensemble learning is performed on the back-end ensemble classifier. The training is iterated until the prediction error is less than a preset error threshold, and a fully trained BERT model is obtained.

[0071] In this optional embodiment, the social text training dataset for each story node's BERT model is an independent training dataset. Samples with corresponding story label semantic information are labeled as positive samples (denoted as 1), and samples without corresponding story label semantic information are labeled as negative samples (denoted as 0). Further, each sample carrying a category label is input into the BERT model for encoding, and a predicted classification result is output through a backend ensemble classifier. This predicted classification result includes the predicted classification and classification probability. The predicted classification result is compared with the corresponding sample's category label, and the prediction error is determined according to a preset loss function. The model parameters of the BERT model are adjusted based on the prediction error, and gradient descent learning is performed on the encoding part of the BERT model. Ensemble learning is performed on the backend ensemble classifier, and iterative training is conducted until the prediction error is less than a preset error threshold, resulting in a fully trained BERT model. In this embodiment, the preset loss function can be any one of the following: exponential loss function, absolute value loss function, perceptual loss function, logarithmic loss function, and squared loss function. The preset error threshold can be within 5%, for example, 2%, 3%, or 4%, and can be set according to actual conditions.

[0072] In this way, by training a complete BERT model for each story node, the matching probability between social text data and each story node and the matching probability with scene lines are determined. Then, by sorting, the N first story lines that best match the social text data are determined, which correspond to the recommendation scenes of K recommended products. The values ​​of N and K can be the same positive integer or different positive integers.

[0073] Step S12: Determine the first lexical vector of each word segment in the social text data, and determine the second lexical vector of the preset keyword of each story tag in each first storyline.

[0074] In an optional embodiment, determining the first lexical vector of each word segment in the social text data, and determining the second lexical vector of the preset keywords of each story tag in each first storyline, includes:

[0075] The social text data is segmented into words, and each segmented word is converted into a corresponding first word vector using the preset word vector model Word2vec.

[0076] Using the pre-defined word vector model Word2vec, all pre-defined keywords of each story tag in the first storyline are converted into corresponding second word vectors.

[0077] Word2vec, a word vector model, is a tool that computationally converts words into vector form. Furthermore, through training, word2vec simplifies text processing into vector operations in a vector space, and the similarity in this vector space can be used to represent the semantic similarity of the text. Therefore, the word vectors output by word2vec can be used for tasks such as clustering, finding synonyms, and part-of-speech tagging, and word2vec is highly efficient.

[0078] In this optional embodiment, a dictionary-based word segmentation algorithm or a statistical machine learning algorithm can be used to segment the social text data. Then, a preset word vector model Word2vec is used to convert each segmented word into a corresponding first word vector. Using the preset word vector model Word2vec, all preset keywords of each story tag in the first storyline are converted into corresponding second word vectors.

[0079] This facilitates subsequent semantic matching. Even if the words in the social text data and the keywords in the keyword library corresponding to the story tags are not exactly the same, they can still be matched if they are semantically similar, thereby improving the success rate of text matching. In addition, compared with traditional keyword libraries, the keyword library for each story tag in this embodiment only needs to store a small number of representative keywords for semantic matching, thereby reducing the number of keywords used.

[0080] Step S13: Determine the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline.

[0081] In an optional embodiment, determining the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline includes:

[0082] Using the cosine similarity metric, the shortest distance from each first word vector of the social text data to all second word vectors of the first storyline is determined, and the average of the shortest distances corresponding to all first word vectors is taken as the similarity between the social text data and the first storyline.

[0083] In this optional embodiment, the word vector model word2vec can be trained for vector operations in the vector space. The calculation logic of the word vector model word2vec is to use the cosine similarity metric to determine the similarity between the social text data and each first storyline. The calculation formula is as follows:

[0084]

[0085] Among them, word i Let keyword be the first word vector of the i-th word segment in the social text data. j Let I be the second word vector of the j-th keyword that has the shortest distance to the first word vector of the i-th word segment in the keyword library of the current first storyline, where I is the number of words in the social text data. Distance represents the graph distance, or similarity, between the current first storyline and the social text data, with a value ranging from 0 to 1.

[0086] In this way, the similarity between each first storyline and social text data can be effectively determined, and the operation is fast even with large amounts of data, based on full matrix operation.

[0087] Step S14: The first storyline whose similarity meets the preset threshold requirement is taken as the target storyline, and the target recommended product is determined based on the target storyline.

[0088] Specifically, the similarity of each first storyline determined in step S13 is sorted from highest to lowest. Then, the first storyline that meets the preset threshold requirement is determined as the target storyline, thereby determining the target recommended product. The preset threshold can be any value between 0.9 and 1, and can be set according to the actual situation.

[0089] As can be seen from the above technical solution, this application constructs multiple scene lines for identifying user needs, then performs semantic matching between the user's social text data and each story node of each scene line to select the most suitable first story line. Then, the social text data is segmented, and each segmented word and each story tag keyword of the first story line are accurately filtered using word vector measurement to obtain the target story line and determine the target recommended product, thereby effectively improving the accuracy of product recommendation and helping to improve the marketing efficiency of subsequent marketing business.

[0090] Please see Figure 2 , Figure 2 This is a functional block diagram of a preferred embodiment of the intelligent product recommendation device based on social data of this application. The intelligent product recommendation device 11 based on social data includes a scene construction unit 110, a data matching unit 111, a vector determination unit 112, a similarity calculation unit 113, and a product recommendation unit 114.

[0091] A scenario unit 110 is constructed to construct multiple scenario lines for identifying user needs. Each scenario line includes several story nodes, each story node includes a story tag and a preset semantic classification model, and each scenario line corresponds to at least one recommended product.

[0092] The matching data unit 111 is used to acquire the user's social text data, determine the matching degree between the social text data and each scene line according to the semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line.

[0093] The vector unit 112 is used to determine the first lexical vector of each word segment in the social text data, and to determine the second lexical vector of the preset keyword of each story tag in each first storyline.

[0094] The similarity calculation unit 113 is used to determine the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline.

[0095] The product recommendation unit 114 is used to take the first storyline whose similarity meets the preset threshold requirement as the target storyline, and to determine the target recommended product based on the target storyline.

[0096] As can be seen from the above technical solution, this application constructs multiple scene lines that identify user needs, then performs semantic matching between the user's social text data and each story node of each scene line to select the most suitable first story line. Then, the social text data is segmented, and each segmented word and the story tag keywords of each first story line are accurately filtered through word vector measurement to obtain the target story line and determine the target recommended product, thereby effectively improving the accuracy of product recommendation and helping to improve the marketing efficiency of subsequent marketing business.

[0097] Specific limitations on each step of the social data-based intelligent product recommendation device can be found in the limitations of the social data-based intelligent product recommendation method described above, and will not be repeated here. Furthermore, it should be noted that each module in the aforementioned social data-based intelligent product recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0098] Please see Figure 3This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 1 includes a memory 12 and a processor 13. The memory 12 is used to store computer-readable instructions, and the processor 13 executes the computer-readable instructions stored in the memory to implement the intelligent product recommendation method based on social data described in any of the above embodiments.

[0099] In an alternative embodiment, the electronic device 1 further includes a bus and a computer program stored in the memory 12 and executable on the processor 13, such as a smart product recommendation program based on social data.

[0100] Figure 3 Only electronic device 1 with memory 12 and processor 13 is shown. It will be understood by those skilled in the art that... Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0101] Combination Figure 1 The memory 12 in the electronic device 1 stores a plurality of computer-readable instructions to implement a smart product recommendation method based on social data, and the processor 13 can execute the plurality of instructions to achieve:

[0102] Construct multiple scene lines for identifying user needs. Each scene line includes several story nodes, each story node includes a story tag and a preset semantic classification model, and each scene line corresponds to at least one recommended product.

[0103] Obtain the user's social text data, determine the matching degree between the social text data and each scene line based on the semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line;

[0104] Determine the first lexical vector for each word segment in the social text data, and determine the second lexical vector for each preset keyword of each story tag in each first storyline;

[0105] The similarity between the social text data and the first storyline is determined based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline.

[0106] The first storyline whose similarity meets the preset threshold requirement is taken as the target storyline, and the target recommended product is determined based on the target storyline.

[0107] Specifically, the processor 13's implementation method for the above instructions can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0108] Those skilled in the art will understand that the schematic diagram is merely an example of electronic device 1 and does not constitute a limitation on electronic device 1. Electronic device 1 can be a bus-type structure or a star-type structure. Electronic device 1 may also include more or fewer other hardware or software than shown in the diagram, or different component arrangements. For example, electronic device 1 may also include input / output devices, network access devices, etc.

[0109] It should be noted that electronic device 1 is only an example. Other existing or future electronic products that are suitable for this application should also be included within the scope of protection of this application and are incorporated herein by reference.

[0110] The memory 12 includes at least one type of readable storage medium, which can be non-volatile or volatile. The readable storage medium includes flash memory, portable hard drives, multimedia cards, card-type memory (e.g., SD or DX memory), magnetic storage, magnetic disks, optical disks, etc. In some embodiments, the memory 12 can be an internal storage unit of the electronic device 1, such as a portable hard drive of the electronic device 1. In other embodiments, the memory 12 can also be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 1. The memory 12 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of a smart product recommendation program based on social data, but also to temporarily store data that has been output or will be output.

[0111] In some embodiments, the processor 13 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 13 is the control unit of the electronic device 1, connecting various components of the electronic device 1 via various interfaces and lines. It executes programs or modules stored in the memory 12 (e.g., executing a smart product recommendation program based on social data) and calls data stored in the memory 12 to perform various functions and process data in the electronic device 1.

[0112] The processor 13 executes the operating system of the electronic device 1 and various installed applications. The processor 13 executes these applications to implement the steps in the various embodiments of the intelligent product recommendation method based on social data described above, for example... Figure 1 The steps are shown.

[0113] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete this application. The one or more modules / units may be a series of computer-readable instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device 1. For example, the computer program may be divided into a scene construction unit 110, a data matching unit 111, a vector determination unit 112, a similarity calculation unit 113, and a product recommendation unit 114.

[0114] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute portions of the intelligent product recommendation method based on social data described in the various embodiments of this application.

[0115] If the modules / units integrated in electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware devices. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above.

[0116] The computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory, and other memory.

[0117] Furthermore, the computer-readable storage medium may primarily include a stored program area and a stored data area, wherein the stored program area may store the operating system, an application program required for at least one function, etc.; and the stored data area may store scene lines, keywords of story tags, and recommended products, etc.

[0118] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, in... Figure 3 The symbol is represented by only one arrow, but this does not indicate that there is only one bus or one type of bus. The bus is configured to enable communication between the memory 12 and at least one processor 13, etc.

[0119] This application also provides a computer-readable storage medium (not shown) storing computer-readable instructions, which are executed by a processor in an electronic device to implement the intelligent product recommendation method based on social data described in any of the above embodiments.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0121] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0122] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0123] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices described in the specification may also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A method for intelligent product recommendation based on social data, characterized in that, The method includes: Construct multiple scene lines for identifying user needs. Each scene line includes several story nodes. Each story node includes a story tag with semantic information and a preset semantic classification model. Each scene line corresponds to at least one recommended product. Each story tag has at least one preset keyword. Obtain the user's social text data, determine the matching degree between the social text data and each scene line based on the semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line; The social text data is segmented into words, and each word in the social text data is converted into a corresponding first word vector using a preset word vector model. The preset word vector model is also used to convert all preset keywords of each story tag in each first storyline into a corresponding second word vector. The similarity between the social text data and the first storyline is determined based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline. The first storyline whose similarity meets the preset threshold requirement is taken as the target storyline, and the target recommended product is determined based on the target storyline; The semantic classification model includes a binary classification model; determining the matching degree between the social text data and each scene line based on the semantic classification model of the multiple scene lines includes: The social text data is input into each of the binary classification models of the scene line for classification, and the matching probability output by each binary classification model is obtained and used as the matching probability of the corresponding story node. Perform a joint probability distribution operation on the matching probabilities of all story nodes in the scene line to obtain the matching probability between the scene line and the social text data, and use it as the matching degree. The step of determining the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline includes: Using the cosine similarity metric, the shortest distance from each first word vector of the social text data to all second word vectors of the first storyline is determined, and the average of the shortest distances corresponding to all first word vectors is taken as the similarity between the social text data and the first storyline.

2. The intelligent product recommendation method based on social data as described in claim 1, characterized in that, The binary classification model includes a BERT model and an ensemble classifier connected to the backend of the BERT model. The matching probability of the story node is the average of the classification probabilities output by all sub-classifiers in the ensemble classifier.

3. The intelligent product recommendation method based on social data as described in claim 2, characterized in that, The training process of the BERT model includes: Obtain a social text training dataset, wherein the samples in the social text training dataset are labeled with positive / negative sample categories; The social text training dataset is input into the BERT model for encoding, and the predicted classification result is output through the backend ensemble classifier. The predicted classification result is compared with the positive / negative sample category of the corresponding sample to determine the prediction error. The model parameters of the BERT model are adjusted according to the prediction error, and gradient descent learning is performed on the encoding part of the BERT model. Ensemble learning is performed on the back-end ensemble classifier. The training is iterated until the prediction error is less than a preset error threshold, and a fully trained BERT model is obtained.

4. The intelligent product recommendation method based on social data as described in claim 1, characterized in that, The preset word vector model is Word2vec.

5. The intelligent product recommendation method based on social data as described in claim 1, characterized in that, The recommended products include at least one service product in at least one financial theme.

6. A smart product recommendation device based on social data, characterized in that, The device includes: A scenario unit is constructed to build multiple scenario lines for identifying user needs. Each scenario line includes several story nodes. Each story node includes a story tag with semantic information and a preset semantic classification model. Each scenario line corresponds to at least one recommended product. Each story tag has at least one preset keyword. The matching data unit is used to acquire the user's social text data, determine the matching degree between the social text data and each scene line according to the semantic classification model of the multiple scene lines, and take the scene line that meets the matching requirements as the first story line. A vector unit is defined to perform word segmentation on the social text data, convert each word segment in the social text data into a corresponding first word vector using a preset word vector model, and convert all preset keywords of each story tag in each first storyline into a corresponding second word vector using the preset word vector model. The similarity calculation unit is used to determine the similarity between the social text data and the first storyline based on all first word vectors corresponding to the social text data and all second word vectors corresponding to each first storyline. The product recommendation unit is used to take the first storyline whose similarity meets the preset threshold requirement as the target storyline, and to determine the target recommended product based on the target storyline; The semantic classification model includes a binary classification model; the matching data unit is further used to input the social text data into each of the binary classification models of the scene line for classification, to obtain the matching probability output by each binary classification model and use it as the matching probability of the corresponding story node, and to perform a joint probability distribution operation on the matching probabilities of all story nodes of the scene line to obtain the matching probability of the scene line and the social text data and use it as the matching degree. The similarity calculation unit is further configured to use the cosine similarity metric to determine the shortest distance from each first word vector of the social text data to all second word vectors of the first storyline, and to use the average of the shortest distances corresponding to all first word vectors as the similarity between the social text data and the first storyline.

7. An electronic device, characterized in that, The electronic device includes: Memory, which stores computer-readable instructions; and The processor executes computer-readable instructions stored in the memory to implement the intelligent product recommendation method based on social data as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the intelligent product recommendation method based on social data as described in any one of claims 1 to 5.

Images