Article information recommendation method and device, network equipment and storage medium
By using recurrent neural network models of session-level, block-level and user-level submodels, combined with micro and macro analysis, the problem of inaccurate recommendation of insurance products is solved, and the accuracy and service quality of user preference determination are improved.
Patent Information
- Application Number
- CN202411907645.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-23
- Publication Date
- 2025-04-25
AI Technical Summary
The lack of fine-grained user behavior analysis in the prior art has led to inaccurate recommendations of insurance products and poor service quality.
A recurrent neural network model including session-level submodel, block-level submodel and user-level submodel is adopted to analyze the user's session sequence information in preferences, combine micro and macro analysis to obtain user's preference information and recommend target items.
It improves the accuracy of user preference determination, enhances the accuracy of item information recommendation, and improves service quality.
Smart Images
Figure CN119884469A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of computer technology, and in particular to an item information recommendation method, device, network equipment and storage medium. Background Art
[0002] With the development of science and technology, more and more services are provided to users, and improving service quality has become the focus of users. Among them, insurance products can be directly analyzed by analyzing user behavior of insurance products. For example, it can be scored by whether the insurance product is clicked or by the rating of the insurance product. However, in the process of analyzing user behavior, there is a lack of fine-grained behavior analysis, which makes the insurance product recommendation inaccurate and the service quality poor. Summary of the invention
[0003] The present disclosure provides an item information recommendation method, device, network device and storage medium to improve the accuracy of item information recommendation and improve service quality. The technical solution of the present disclosure is as follows: According to a first aspect of an embodiment of the present disclosure, there is provided a method for recommending item information, comprising: Acquire multi-period interaction information corresponding to the target user identifier, wherein the multi-period interaction information includes a session sequence information set corresponding to the item set; Using a first recurrent neural network model, performing preference analysis on at least one session sequence information in the session sequence information set to obtain preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; According to the preference information, target item information is recommended to the target user identifier.
[0004] According to some embodiments, the acquiring multi-period interaction information corresponding to the target user identifier includes: Acquire operation behavior information corresponding to the target user identifier in each period of multiple periods, wherein the operation behavior information includes a preset operation behavior for a target item, a time interval of the same preset operation, and a residence time of the target item; According to the operation behavior information and the representation method corresponding to the operation behavior, the interaction information corresponding to the target user identifier is acquired.
[0005] According to some embodiments, the adopting the first recurrent neural network model to perform preference analysis on the at least one session sequence information to obtain preference information corresponding to the target user identifier includes: Obtaining a subset of the session sequence information corresponding to each historical moment in the session sequence information set; Using a session-level sub-model to identify a subset of session sequence information corresponding to each of the historical moments, and obtaining a first preference representation corresponding to each of the historical moments; Acquire a first preference representation set corresponding to any period in at least one period, wherein the first preference representation set includes first preference representations corresponding to each historical moment in any period, and the any period includes at least one historical moment; According to the time information of each first preference representation in the first preference representation set, the first preference representation set is processed by using a block-level sub-model to obtain a second preference representation corresponding to any one of the periods; Acquire a second preference representation set corresponding to the multi-period; According to the time information of each second preference representation in the second preference representation set, the second preference representation set is processed using a user-level sub-model to obtain a third preference representation, and the third preference representation is used as the preference information corresponding to the target user identifier.
[0006] According to some embodiments, the session-level sub-model includes an input gate, a forget gate, a time gate, a behavior gate, and an output gate, and the use of the session-level sub-model to identify the subset of session sequence information corresponding to each historical moment to obtain the first preference representation corresponding to each historical moment includes: The input gate, the forget gate, the time gate, the behavior gate and the output gate are respectively used to process the subset of the session sequence information of each historical moment to obtain the output representation corresponding to each gate at each historical moment; The first preference representation of each historical moment is obtained according to the output representation corresponding to each gate, the weight matrix corresponding to each gate, and the bias information corresponding to each gate.
[0007] According to some embodiments, the obtaining the second preference representation corresponding to the any period includes: Obtaining a previous third preference representation corresponding to any one of the cycles; A fusion process is performed on the last third preference representation corresponding to any one period and the first preference representation set to obtain a second preference representation corresponding to any one period.
[0008] According to some embodiments, the method further comprises: According to the fusion layer parameters corresponding to each sub-model in the first recurrent neural network model, the preference representation corresponding to each sub-model is updated to obtain an updated preference representation.
[0009] According to some embodiments, the method further comprises: Get the training data set; The loss function is constructed using Bayesian personalized ranking BPR-Max; The second recurrent neural network model is trained using the training data set and the loss function, and the first recurrent neural network model is obtained when the output result of the second recurrent neural network model meets the result requirements.
[0010] According to a second aspect of an embodiment of the present disclosure, there is provided an item information recommendation device, comprising: A set acquisition unit, configured to acquire multi-period interaction information corresponding to a target user identifier, wherein the multi-period interaction information includes a session sequence information set corresponding to an item set; an information acquisition unit, configured to perform preference analysis on at least one piece of session sequence information in the session sequence information set using a first recurrent neural network model, and acquire preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; An information recommendation unit is used to recommend target item information to the target user identifier according to the preference information.
[0011] According to a third aspect of an embodiment of the present disclosure, a network device is provided, including: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the item information recommendation method described in any one of the aforementioned aspects.
[0012] According to a fourth aspect of an embodiment of the present disclosure, a storage medium is provided. When instructions in the storage medium are executed by a processor of a network device, the network device is enabled to execute the item information recommendation method described in any one of the preceding aspects.
[0013] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program implements any one of the methods described in the preceding aspects when executed by a processor.
[0014] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects: In some or related embodiments, by acquiring multi-period interaction information corresponding to a target user identifier, wherein the multi-period interaction information includes a conversation sequence information set corresponding to a set of items; using a first recurrent neural network model, performing preference analysis processing on at least one conversation sequence information in the conversation sequence information set, obtaining preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a conversation-level sub-model, a block-level sub-model, and a user-level sub-model; and recommending target item information to the target user identifier according to the preference information. Therefore, the user's conversation sequence can be analyzed by a model including a conversation-level sub-model, a block-level sub-model, and a user-level sub-model, reducing the situation where item recommendations are inaccurate based only on the user's macro-behavior analysis, combining the micro-analysis and macro-analysis collection method, and recommending items by combining the interaction information of the user identifier and the preference representation, obtaining the user identifier's preference dynamic change information, which can improve the accuracy of user preference determination, improve the accuracy of item information recommendation, and improve service quality.
[0015] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.
[0017] Figure 1 is a flow chart of a method for recommending item information according to an exemplary embodiment; Figure 2 is a flow chart of a method for recommending item information according to an exemplary embodiment; Figure 3 is a flow chart of a method for recommending item information according to an exemplary embodiment; Figure 4 is a specific flow chart of a method for recommending item information according to an exemplary embodiment; Figure 5 is a symbolic illustration of a specific flow chart of a method for recommending item information according to an exemplary embodiment; Figure 6 is a block diagram of an item information recommendation device according to an exemplary embodiment; Figure 7 The invention is a block diagram of a network device according to an exemplary embodiment. DETAILED DESCRIPTION
[0018] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.
[0019] The embodiments of the present disclosure provide an item information recommendation method, device, network device, and storage medium. In some embodiments, the item information recommendation method, information processing method, communication method, and other terms can be replaced with each other, the item information recommendation device, information processing device, communication device, and other terms can be replaced with each other, and the information processing system, communication system, and other terms can be replaced with each other.
[0020] The embodiments of the present disclosure are not exhaustive, but are only illustrative of some embodiments, and are not intended to be a specific limitation on the scope of protection of the present disclosure. In the absence of contradiction, each step in a certain embodiment can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a certain embodiment can also be implemented as an independent embodiment, and the order of the steps in a certain embodiment can be arbitrarily exchanged. In addition, the optional implementation methods in a certain embodiment can be arbitrarily combined; in addition, the embodiments can be arbitrarily combined, for example, some or all of the steps of different embodiments can be arbitrarily combined, and a certain embodiment can be arbitrarily combined with the optional implementation methods of other embodiments.
[0021] In each embodiment of the present disclosure, unless otherwise specified or there is a logical conflict, the terms and / or descriptions between the embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form a new embodiment based on their internal logical relationships.
[0022] The terms used in the embodiments of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure.
[0023] In the embodiments of the present disclosure, unless otherwise specified, elements expressed in the singular form, such as "a", "an", "the", "above", "said", "aforementioned", "this", etc., may mean "one and only one", or "one or more", "at least one", etc. For example, when using articles such as "a", "an", "the" in English in translation, the noun after the article may be understood as a singular expression or a plural expression.
[0024] In the embodiments of the present disclosure, “plurality” refers to two or more.
[0025] In some embodiments, the terms “at least one,” “one or more,” “a plurality of,” “multiple,” etc. may be used interchangeably.
[0026] In some embodiments, "at least one of A and B", "A and / or B", "A in one case, B in another case", "in response to one case A, in response to another case B", etc., may include the following technical solutions according to the situation: in some embodiments, A (A is executed independently of B); in some embodiments, B (B is executed independently of A); in some embodiments, execution is selected from A and B (A and B are selectively executed); in some embodiments, A and B (A and B are both executed). When there are more branches such as A, B, C, etc., the above is also similar.
[0027] In some embodiments, devices and equipment may be interpreted as physical or virtual, and their names are not limited to the names recorded in the embodiments. In some cases, they may also be understood as "equipment", "device", "circuit", "network element", "node", "function", "unit", "section", "system", "network", "chip", "chip system", "entity", "subject", etc.
[0028] In some embodiments, the term “terminal” or “terminal device” may be referred to as “user equipment (UE)”, “user terminal”, “mobile station (MS)”, “mobile terminal (MT)”, subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, etc.
[0029] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.
[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0031] Figure 1 is a flow chart of a method for recommending item information according to an exemplary embodiment. Figure 1 As shown, the following steps are included: In step S11, multi-period interaction information corresponding to the target user identifier is obtained, wherein the multi-period interaction information includes a conversation sequence information set corresponding to the item set; In some embodiments, the execution subject of the embodiments of the present disclosure may be, for example, a network device, and the name of the network device is not limited. For example, the network device may also be called a server, a base station, or a server cluster, etc. The embodiments of the present disclosure are not limited to this. The technical solution of the embodiments of the present disclosure may be applied to, for example, an insurance product recommendation scenario.
[0032] According to some embodiments, the user identifier is used to uniquely identify the user, that is, different users correspond to different user identifiers. There is no limitation on the composition of the user identifier. The composition of the user identifier includes but is not limited to letters, numbers, symbols, etc.
[0033] In some embodiments, the target user identifier is used to indicate the user identifier of the item information to be recommended. The target user identifier does not specifically refer to a fixed user identifier. When a modification instruction for the target user identifier is received, the target user identifier may also change accordingly. For example, when the user identifier composition method changes, the target user identifier may also change accordingly.
[0034] In some embodiments, a multi-period may be used to indicate a plurality of historical periods, for example, and the multi-period may be a plurality of periods determined according to a preset period duration. The multi-period does not specifically refer to a fixed period, for example, when the number of periods corresponding to the multi-period changes, the multi-period may also change accordingly. For example, when the period duration corresponding to each period changes, the multi-period may also change accordingly.
[0035] According to some embodiments, the interaction information may be used to indicate information about the target user identification interacting with each item in the item set within a multi-period. The interaction information does not specifically refer to a fixed set. For example, when the multi-period changes, the interaction information may also change accordingly. For example, when the item set changes, the interaction information may also change accordingly.
[0036] In some embodiments, the item set is, for example, a collection of at least one item. The item set does not specifically refer to a fixed set. For example, when the number of items included in the item set changes, the item set may also change accordingly. For example, when an item in the item set changes, the item set may also change accordingly.
[0037] In some embodiments, multi-period interaction information corresponding to the target user identifier is obtained, wherein the multi-period interaction information includes a set of session sequence information corresponding to the set of items. The session sequence information may be information constructed in time according to the behavior sequence of the user identifier.
[0038] In step S12, a first recurrent neural network model is used to perform preference analysis on at least one session sequence information in the session sequence information set to obtain preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; According to some embodiments, a recurrent neural network (RNN) model is a large-model neural network that can process time series information.
[0039] In some embodiments, the first recurrent neural network model may be, for example, a model that has been trained and can perform item recommendations. The first recurrent neural network model is used to distinguish it from other models and does not specifically refer to a fixed model. For example, when the number of sub-models included in the first recurrent neural network model changes, the first recurrent neural network model may also change accordingly.
[0040] In some embodiments, the preference information is used to indicate the preference information corresponding to the target user identifier in the multi-cycle. The preference information does not specifically refer to a fixed information. For example, when the multi-cycle changes, the preference information may also change accordingly. For example, when the first recurrent neural network model changes, the preference information may also change accordingly.
[0041] In some embodiments, the submodel may be used to indicate a submodel included in a first recurrent neural network model. The first recurrent neural network model may include, for example, a session-level submodel, a block-level submodel, and a user-level submodel. The session-level submodel, the block-level submodel, and the user-level submodel may be used to determine preference changes of a target user identifier in multiple cycles.
[0042] In some embodiments, a first recurrent neural network model is used to perform preference analysis on at least one session sequence information in the session sequence information set to obtain preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model.
[0043] In step S13, target item information is recommended to the target user identifier according to the preference information.
[0044] In some embodiments, the target item information to be recommended in the item set can be obtained according to the preference information, and the target item information can be recommended to the target user identifier. The display method of the target item information is not limited. For example, the name of the target item information can be directly displayed, the animation information corresponding to the target item information can be displayed, or the target item information can be displayed by voice.
[0045] In some or related embodiments, by acquiring multi-cycle interaction information corresponding to the target user identifier, wherein the multi-cycle interaction information includes a conversation sequence information set corresponding to the item set; using a first recurrent neural network model, performing preference analysis processing on at least one conversation sequence information in the conversation sequence information set, obtaining preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a conversation level sub-model, a block level sub-model and a user level sub-model; and recommending target item information to the target user identifier according to the preference information. Therefore, the user's conversation sequence can be analyzed by a model including a conversation level sub-model, a block level sub-model and a user level sub-model, reducing the situation where the item recommendation is inaccurate based only on the macro behavior analysis of the user, and combining the micro analysis and macro analysis to recommend items by combining the interaction information of the user identifier and the preference expression, obtaining the dynamic change information of the user identifier's preference, which can improve the accuracy of user preference determination, improve the accuracy of item information recommendation, and improve the service quality. In addition. The recurrent neural network model may include a session-level sub-model, a block-level sub-model, and a user-level sub-model, and may jointly model multi-period preference changes of user identifiers, thereby improving the accuracy of preference determination and the accuracy of item recommendations.
[0046] Figure 2 is a flow chart of a method for recommending item information according to an exemplary embodiment. Figure 2 As shown, the following steps are included: In step S21, the operation behavior information corresponding to the target user identifier in each cycle of the multiple cycles is obtained, wherein the operation behavior information includes a preset operation behavior for the target item, a time interval of the same preset operation, and a residence time of the target item; The specific process is as described above and will not be repeated here.
[0047] In some embodiments, the technical solution of the disclosed embodiment can be applied to the item recommendation scenario, for example, the recommendation of insurance products. Therefore, the technical solution of the disclosed embodiment can integrate the preference learning in the user's micro-operation details with the multi-period preference evolution of the macro-session sequence, and build a more accurate and comprehensive user preference model to serve the precise and personalized insurance product intelligent recommendation service.
[0048] According to some embodiments, the operation behavior information is used to indicate the operation behavior corresponding to each period of the multi-period. The operation behavior information may be for a set of items, that is, information for operations on multiple items may be obtained. The preset operation behavior may include, for example, a search operation, a review operation, a favorite operation, an add-to-cart operation, a purchase operation, etc.
[0049] In some embodiments, the operation behavior information corresponding to the target user identifier in each period of multiple periods can be obtained, wherein the operation behavior information includes preset operation behaviors for the target item, the time interval of the same preset operation, and the residence time of the target item.
[0050] In step S22, according to the operation behavior information and the representation method corresponding to the operation behavior, the interaction information corresponding to the target user identifier is obtained, wherein the multi-period interaction information includes a conversation sequence information set corresponding to the item set; The specific process is as described above and will not be repeated here.
[0051] According to some embodiments, Figure 3 is a flowchart of a method for recommending item information according to an exemplary embodiment. Figure 4 is a specific flow chart of a method for recommending item information according to an exemplary embodiment. Figure 4 As shown, different sub-models are described by different colors, and Figure 4 The description of each symbol is as follows Figure 5 As shown in Figure 1, the item set can be called an insurance product item set, for example. Therefore, the recurrent neural network model can be as follows: Figure 4 The left half of the figure starts with the micro-operations in a conversation sequence to gain insight into the fine-grained preferences of users. On this basis, the right half of the figure learns the multi-period preference representation and evolution in the user conversation sequence from a macro perspective. To this end, a hierarchical RNN structure is established, including: session level, block level and user level, to learn and integrate the short-term, medium-term and long-term preference representations of users respectively. The model integrates the two to construct an accurate and comprehensive user preference model UIP.
[0052] like Figure 4As shown in the figure, from a micro perspective, for the fine-grained user preferences implied by the interactive behaviors between user identifiers and insurance products in a session, such as adding to favorites, dwell time, etc., an RNN network with two behavior-gated LSTMses structures is designed to learn the fine-grained immediate preferences of users in multiple behavior types. From a macro perspective, for multi-period session sequences: recent session-based, mid-term block-based, and long-term user-based session sequences, a recurrent neural network RNN is obtained to learn the dynamic evolution of users' short-term, mid-term, and long-term preferences.
[0053] According to some embodiments, the user identification set may include, for example, U=u1, u2, ..., um; the insurance product item set includes V=v1, v2, ..., vn; wherein m and n represent the number of user identifications and insurance product items, respectively. After arranging the conversation sequence between user u and the insurance product item set by time, the conversation block set of u is located as follows: , where nB represents the number of session blocks generated by user u. , which is composed of a series of sessions, where ns indicates the number of user sessions in block i, and the jth session It is composed of a series of operation behaviors, where xs represents the number of behaviors in the session.
[0054] According to some embodiments, bk1: search operation; bk2: review operation; bk3: favorite operation; bk4: add to cart operation; bk5: purchase operation. The operation behavior xk in can further locate the five-tuple expression form, where is the micro-operation behavior, and xk is the total embedding vector representation.
[0055] In step S23, a first recurrent neural network model is used to perform preference analysis on at least one session sequence information in the session sequence information set to obtain preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; The specific process is as described above and will not be repeated here.
[0056] According to some embodiments, the adopting the first recurrent neural network model to perform preference analysis on the at least one session sequence information to obtain preference information corresponding to the target user identifier includes: Obtaining a subset of the session sequence information corresponding to each historical moment in the session sequence information set; Using a session-level sub-model to identify a subset of session sequence information corresponding to each of the historical moments, and obtaining a first preference representation corresponding to each of the historical moments; Acquire a first preference representation set corresponding to any period in at least one period, wherein the first preference representation set includes first preference representations corresponding to each historical moment in any period, and the any period includes at least one historical moment; According to the time information of each first preference representation in the first preference representation set, the first preference representation set is processed by using a block-level sub-model to obtain a second preference representation corresponding to any one of the periods; Acquire a second preference representation set corresponding to the multi-period; According to the time information of each second preference representation in the second preference representation set, the second preference representation set is processed using a user-level sub-model to obtain a third preference representation, and the third preference representation is used as the preference information corresponding to the target user identifier.
[0057] According to some embodiments, the session-level sub-model includes an input gate, a forget gate, a time gate, a behavior gate, and an output gate, and the use of the session-level sub-model to identify the subset of session sequence information corresponding to each historical moment to obtain the first preference representation corresponding to each historical moment includes: The input gate, the forget gate, the time gate, the behavior gate and the output gate are respectively used to process the subset of the session sequence information of each historical moment to obtain the output representation corresponding to each gate at each historical moment; The first preference representation of each historical moment is obtained according to the output representation corresponding to each gate, the weight matrix corresponding to each gate, and the bias information corresponding to each gate.
[0058] In some embodiments, the session-level sub-model may include, for example, an input gate, a forget gate, a time gate, a behavior gate, and an output gate, wherein the input gate, the forget gate, and the time gate are used to process the behavior type and the dwell time of the target user representation in the session. Among them, the behavior gate can process the behavior of the user identifier, and the session-level sub-model can obtain the session-level short-term behavior representation carrying the user behavior details. The specific working process of the session-level sub-model is as follows: ; ; ; ; ; ; ; ;in, are the output representations of the input gate, forget gate, time gate, behavior gate, and output gate of the behavior at time t, respectively. C and h are the cell state and hidden state vectors in the classic LSTM structure, respectively. are the corresponding weight matrices, are the corresponding biases respectively. The input to the structure is a triple ( ),in: is the embedding vector of user’s behavior at time t; is the embedding vector of the behavior type; Is the current behavior With the next behavior The time interval between embedding vectors.
[0059] According to some embodiments, obtaining a second preference representation corresponding to any one of the cycles includes: Obtaining a previous third preference representation corresponding to any one of the cycles; The last third preference representation corresponding to any one of the cycles and the first preference representation set are merged to obtain a second preference representation corresponding to any one of the cycles. The last third preference representation corresponding to any one of the cycles may be, for example, a third preference representation corresponding to a previous multi-cycle of the multi-cycle in which any one of the cycles is located.
[0060] According to some embodiments, the method further comprises: According to the fusion layer parameters corresponding to each sub-model in the first recurrent neural network model, the preference representation corresponding to each sub-model is updated to obtain an updated preference representation.
[0061] In some embodiments, the session-level submodel is a session-level , the block-level sub-model is block-level , the user-level submodel is user-level , the hidden state is the preference representation. Session level The initial state of the previous moment The hidden state of the output is updated, Medium Conversation The initial hidden state It is expressed as: .in, Is the previous session block At the end The hidden state of the output is the session in the next block Communicates a user's medium-term preference representation. and They are Initialize weights and biases. is the hyperbolic tangent activation function. In addition, Add a fusion layer to explore the user behavior embedding representation The complex interactive relationship between the medium-term preference representation of the previous moment (as shown by the green line in the model diagram) and the user's long-term preference representation (as shown by the red line in the model diagram). This can provide a priori guidance on medium-term and global preference representation for short-term user preference learning. The fusion method used in this device is based on the link and product method, which is publicized as: .
[0062] in, Is the last session block At the end The hidden state of the output is regarded as the user's long-term preference representation; ⊙ is the hadamard product (element product) operation. The left side of the formula is based on the connection-based fusion operation: it is assumed that the input user behavior embedding vector, medium-term preference representation and long-term preference representation are independent of each other, and the information at different levels is directly connected; the right side is based on the product-based fusion operation: it is assumed that there is a close correlation between the information, and the interaction mode of information at different levels can be better captured by the product method; this formula is based on the connection and product fusion operation mode, coordinating the above two assumptions. Finally, the constructed ,in, and is the fused weight matrix and bias vector. Finally, based on the fusion layer and Prior information, through Hidden state of the structure output To update the session-level short-term user preference representation: .
[0063] According to some embodiments, since medium- and long-term behaviors do not need to consider user-level operation behaviors, block-level and user level The LSTM three-gate structure design is adopted. The initial state of the previous moment is The long-term preference representation is updated by initializing the input In the above example, the traditional random initialization setting is mentioned. As shown in formula (1): (1) in, At the end of the last moment Output the user's long-term preference representation. and They are Initialize weights and biases.
[0064] For a conversation sequence , The hidden state of the final output Indicates the user's short-term immediate preference at the last moment. The hidden state of the output In order to effectively integrate the two types of information to guide and update the current Medium-term learning preference, choose to use a concatenation- and product-based approach ] Based on the above fusion mechanism, the final construction The fusion layer : (2) Through the design of the fusion layer, The structure can well model the dynamics between adjacent sessions within a block, thereby learning the evolution of users' mid-term preferences. Output hidden state As an update of the user's mid-term preferences. .
[0065] According to some embodiments, in the long-term user preference learning, the previous moment is integrated and The output of the user's short-term and medium-term preference representation is at the user level A fusion layer is added, and the fusion method is based on connection and product. , and then build based on the fusion mechanism The fusion layer (3), (3) at last, Output hidden state , to represent user-level long-term preferences (4): (4) In some embodiments, the loss function can be constructed using the optimized Bayesian personalized ranking BPR-Max . Optimize the loss function objective to make the recommended items higher than the maximum sample product score ( ) minimizes the negative log probability , for example, it can be as shown in formula (5), (5) in, is the target insurance product item, j is a negative sample randomly sampled from the insurance product item set (a product item that the user has never interacted with); N is the set of negative samples, and is the recommendation score of the model for insurance product items i and j. and is discontinuous, so through the sigmoid function calculate ,get Probability value; at the same time, through Approximate calculation The probability value of ,in, is a regularization term that makes the negative sample score approach 0. is the regularization term coefficient. Finally, the recommendation of insurance product items is given by , and The output of the user's short-term, medium-term and long-term preferences are expressed as: , , Calculated together, .in: , , . Select the exponential linear unit function (ExponentialLinear Units).
[0066] According to some embodiments, It is an evolution of the ReLU activation function, which can effectively avoid the disappearance of gradients. At the same time, it also tries to make the average value of the urgent activation function close to zero, thereby speeding up learning. and To adjust the weight coefficient of block-level and user-level information. Finally, by minimizing the loss function To train the model:
[0067] According to some embodiments, the method further comprises: Get the training data set; The loss function is constructed using Bayesian personalized ranking BPR-Max; The second recurrent neural network model is trained using the training data set and the loss function, and the first recurrent neural network model is obtained when the output result of the second recurrent neural network model meets the result requirements.
[0068] In step S24, target item information is recommended to the target user identifier according to the preference information.
[0069] In one or related embodiments, operational behavior information corresponding to the target user identifier in each period of multiple periods can be obtained, wherein the operational behavior information includes preset operational behaviors for target items, time intervals for the same preset operations, and residence time of the target items; based on the operational behavior information and the representation method corresponding to the operational behaviors, interactive information corresponding to the target user identifier is obtained, wherein the multiple period interactive information includes a session sequence information set corresponding to a set of items; therefore, the session sequence information set can be obtained through a combination of micro and macro, which can improve the accuracy of item recommendations and improve service quality.
[0070] Figure 6 is a block diagram of an item information recommendation device according to an exemplary embodiment. Figure 6 , the item information recommendation device 600 comprises: A set acquisition unit 601 is used to acquire multi-period interaction information corresponding to a target user identifier, wherein the multi-period interaction information includes a session sequence information set corresponding to an item set; An information acquisition unit 602 is used to perform preference analysis on at least one session sequence information in the session sequence information set using a first recurrent neural network model to acquire preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; The information recommendation unit 603 is used to recommend target item information to the target user identifier according to the preference information.
[0071] According to some embodiments, the set acquisition unit 601, when used to acquire multi-period interaction information corresponding to the target user identifier, is specifically used to: Acquire operation behavior information corresponding to the target user identifier in each period of multiple periods, wherein the operation behavior information includes a preset operation behavior for a target item, a time interval of the same preset operation, and a residence time of the target item; According to the operation behavior information and the representation method corresponding to the operation behavior, the interaction information corresponding to the target user identifier is acquired.
[0072] According to some embodiments, the information acquisition unit 602 is used to use the first recurrent neural network model to perform preference analysis on the at least one session sequence information to obtain preference information corresponding to the target user identifier, specifically for: Obtaining a subset of the session sequence information corresponding to each historical moment in the session sequence information set; Using a session-level sub-model to identify a subset of session sequence information corresponding to each of the historical moments, and obtaining a first preference representation corresponding to each of the historical moments; Acquire a first preference representation set corresponding to any period in at least one period, wherein the first preference representation set includes first preference representations corresponding to each historical moment in any period, and the any period includes at least one historical moment; According to the time information of each first preference representation in the first preference representation set, the first preference representation set is processed by using a block-level sub-model to obtain a second preference representation corresponding to any one of the periods; Acquire a second preference representation set corresponding to the multi-period; According to the time information of each second preference representation in the second preference representation set, the second preference representation set is processed using a user-level sub-model to obtain a third preference representation, and the third preference representation is used as the preference information corresponding to the target user identifier.
[0073] According to some embodiments, the session-level sub-model includes an input gate, a forget gate, a time gate, a behavior gate, and an output gate, and the information acquisition unit 602 is used to use the session-level sub-model to identify the session sequence information subset corresponding to each historical moment, and to obtain the first preference representation corresponding to each historical moment, specifically for: The input gate, the forget gate, the time gate, the behavior gate and the output gate are respectively used to process the subset of the session sequence information of each historical moment to obtain the output representation corresponding to each gate at each historical moment; The first preference representation of each historical moment is obtained according to the output representation corresponding to each gate, the weight matrix corresponding to each gate, and the bias information corresponding to each gate.
[0074] According to some embodiments, the information acquisition unit 602, when used to acquire the second preference representation corresponding to any one of the periods, is specifically used to: Obtaining a previous third preference representation corresponding to any one of the cycles; A fusion process is performed on the last third preference representation corresponding to any one period and the first preference representation set to obtain a second preference representation corresponding to any one period.
[0075] According to some embodiments, the information acquisition unit 602 is further specifically configured to: According to the fusion layer parameters corresponding to each sub-model in the first recurrent neural network model, the preference representation corresponding to each sub-model is updated to obtain an updated preference representation.
[0076] According to some embodiments, the information acquisition unit 602 is further specifically configured to: Get the training data set; The loss function is constructed using Bayesian personalized ranking BPR-Max; The second recurrent neural network model is trained using the training data set and the loss function, and the first recurrent neural network model is obtained when the output result of the second recurrent neural network model meets the result requirements.
[0077] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0078] In some or related embodiments, a set acquisition unit 601 is used to acquire multi-period interaction information corresponding to a target user identifier, wherein the multi-period interaction information includes a session sequence information set corresponding to an item set; an information acquisition unit 602 is used to perform preference analysis processing on at least one session sequence information in the session sequence information set using a first recurrent neural network model to acquire preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; an information recommendation unit 603 is used to recommend target item information to the target user identifier according to the preference information. Therefore, the user's session sequence can be analyzed by a model including a session-level sub-model, a block-level sub-model, and a user-level sub-model, thereby reducing the situation where item recommendations are inaccurate based only on the user's macro-behavior analysis. The micro-analysis and macro-analysis collection method can be combined to recommend items by combining the interaction information of the user identifier and the preference representation, and the user identifier's preference dynamic change information can be acquired, thereby improving the accuracy of user preference determination, improving the accuracy of item information recommendation, and improving service quality.
[0079] Figure 7 700 is a block diagram of a network device 700 provided in an embodiment of the present disclosure. For example, the network device 700 may be provided as a network side device. Figure 7The network device 700 includes a processing component 722, which further includes at least one processor, and a memory resource represented by a memory 732 for storing instructions executable by the processing component 722, such as an application. The application stored in the memory 732 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 722 is configured to execute instructions to perform the above-mentioned method.
[0080] The network device 700 may also include a power supply component 727 configured to perform power management of the network device 700, a wired or wireless network interface 750 configured to connect the network device 700 to a network, and an input / output (I / O) interface 758. The network device 700 may operate based on an operating system stored in the memory 732, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM or the like.
[0081] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0082] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable item information recommendation device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.
[0083] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0084] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).
[0085] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.
[0086] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.
[0087] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.
[0088] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A method for recommending item information, characterized in that: include: Acquire multi-period interaction information corresponding to the target user identifier, wherein the multi-period interaction information includes a session sequence information set corresponding to the item set; Using a first recurrent neural network model, performing preference analysis on at least one session sequence information in the session sequence information set to obtain preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; According to the preference information, target item information is recommended to the target user identifier.
2. The method according to claim 1, characterized in that The acquiring of multi-period interaction information corresponding to the target user identifier includes: Acquire operation behavior information corresponding to the target user identifier in each period of multiple periods, wherein the operation behavior information includes a preset operation behavior for a target item, a time interval of the same preset operation, and a residence time of the target item; According to the operation behavior information and the representation method corresponding to the operation behavior, the interaction information corresponding to the target user identifier is acquired.
3. The method according to claim 1, characterized in that The adopting the first recurrent neural network model to perform preference analysis on the at least one session sequence information to obtain preference information corresponding to the target user identifier includes: Obtaining a subset of the session sequence information corresponding to each historical moment in the session sequence information set; Using a session-level sub-model to identify a subset of session sequence information corresponding to each of the historical moments, and obtaining a first preference representation corresponding to each of the historical moments; Acquire a first preference representation set corresponding to any period in at least one period, wherein the first preference representation set includes first preference representations corresponding to each historical moment in any period, and the any period includes at least one historical moment; According to the time information of each first preference representation in the first preference representation set, the first preference representation set is processed by using a block-level sub-model to obtain a second preference representation corresponding to any one of the periods; Acquire a second preference representation set corresponding to the multi-period; According to the time information of each second preference representation in the second preference representation set, the second preference representation set is processed using a user-level sub-model to obtain a third preference representation, and the third preference representation is used as the preference information corresponding to the target user identifier.
4. The method according to claim 3, characterized in that in, The session-level sub-model includes an input gate, a forget gate, a time gate, a behavior gate, and an output gate. The session-level sub-model is used to identify the session sequence information subset corresponding to each historical moment to obtain the first preference representation corresponding to each historical moment, including: The input gate, the forget gate, the time gate, the behavior gate and the output gate are respectively used to process the subset of the session sequence information of each historical moment to obtain the output representation corresponding to each gate at each historical moment; The first preference representation of each historical moment is obtained according to the output representation corresponding to each gate, the weight matrix corresponding to each gate, and the bias information corresponding to each gate.
5. The method according to claim 4, characterized in that The obtaining of a second preference representation corresponding to any one of the cycles includes: Obtaining a previous third preference representation corresponding to any one of the cycles; A fusion process is performed on the last third preference representation corresponding to any one period and the first preference representation set to obtain a second preference representation corresponding to any one period.
6. The method according to claim 4, characterized in that The method further comprises: According to the fusion layer parameters corresponding to each sub-model in the first recurrent neural network model, the preference representation corresponding to each sub-model is updated to obtain an updated preference representation.
7. The method according to claim 1, characterized in that The method further comprises: Get the training data set; The loss function is constructed using Bayesian personalized ranking BPR-Max; The second recurrent neural network model is trained using the training data set and the loss function, and the first recurrent neural network model is obtained when the output result of the second recurrent neural network model meets the result requirements.
8. An item information recommendation device, characterized in that: include: A set acquisition unit, configured to acquire multi-period interaction information corresponding to a target user identifier, wherein the multi-period interaction information includes a session sequence information set corresponding to an item set; an information acquisition unit, configured to perform preference analysis on at least one piece of session sequence information in the session sequence information set using a first recurrent neural network model, and acquire preference information corresponding to the target user identifier, wherein the first recurrent neural network model includes a session-level sub-model, a block-level sub-model, and a user-level sub-model; An information recommendation unit is used to recommend target item information to the target user identifier according to the preference information.
9. A network device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the item information recommendation method according to any one of claims 1 to 7.
10. A storage medium, when instructions in the storage medium are executed by a processor of a network device, the network device is enabled to execute the item information recommendation method according to any one of claims 1 to 7.