Information recommendation methods, devices, computer equipment and storage media

Abstract

The embodiment of the present application belongs to the field of artificial intelligence, and relates to an information recommendation method, device, computer equipment and storage medium, the method comprising: obtaining a question contribution queue composed of questions arranged according to contribution; determining a question sorting exchange combination of the question contribution queue to adjust the sorting thereof to obtain a plurality of first question lists; displaying the first question list and receiving a first operation information for the first question list; calculating the click value score of each question according to each first operation information; inputting the question sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value; calculating the model loss according to the exchange prediction evaluation value and the click value score to adjust the initial exchange prediction model to obtain an exchange prediction model; inputting the question sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value; adjusting the first question list according to the exchange evaluation value to obtain a second question list for information recommendation. The present application improves the efficiency of information recommendation.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an information recommendation method, device, computer equipment and storage medium. Background Art

[0002] With the development of computer technology, online shopping has developed rapidly. People can browse various products through terminals and make purchases. In order to display products, merchants usually create complex and cumbersome pages and introduce products in detail. However, this usually causes users to spend a long time to familiarize themselves with the products, or makes it difficult for users to find information of interest, resulting in low efficiency of information push and reduced user conversion rate. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide an information recommendation method, apparatus, computer equipment and storage medium to solve the problem of low efficiency in pushing product information.

[0004] In order to solve the above technical problems, the present application embodiment provides an information recommendation method, which adopts the following technical solution:

[0005] Obtaining a question contribution queue of a topic node, wherein the topic node is a node in an information tree representing a product information system, and the question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the degree of contribution of the question to product sales;

[0006] Determine a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, and adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists;

[0007] For each first question list, display the first question list through the first terminal, and receive first operation information of the user for the first question list through the first terminal;

[0008] Calculate the click value score of each question in the question contribution queue according to each first operation information;

[0009] Inputting the problem sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value of the problem sorting exchange combination;

[0010] Calculating a model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain an exchange prediction model;

[0011] Inputting the problem sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value;

[0012] The first question list is adjusted according to the obtained exchange evaluation value to obtain a second question list, and the second question list is used to be displayed through the second terminal to achieve information recommendation.

[0013] In order to solve the above technical problems, the embodiment of the present application further provides an information recommendation device, which adopts the following technical solution:

[0014] A queue acquisition module is used to acquire a question contribution queue of a topic node, wherein the topic node is a node in an information tree representing a product information system, and the question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the contribution of the question to product sales;

[0015] A sorting adjustment module, used for determining a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, so as to adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists;

[0016] An information receiving module, configured to display, for each first question list, the first question list through a first terminal, and receive, through the first terminal, first operation information of a user with respect to the first question list;

[0017] A score calculation module, used to calculate the click value score of each question in the question contribution queue according to each first operation information;

[0018] A combination input module, used for inputting the problem sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value of the problem sorting exchange combination;

[0019] A model adjustment module, used for calculating a model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain an exchange prediction model;

[0020] A question input module, used for inputting the question sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value;

[0021] The list adjustment module is used to adjust the first question list according to the obtained exchange evaluation value to obtain a second question list, and the second question list is used to be displayed through the second terminal to realize information recommendation.

[0022] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0023] Obtaining a question contribution queue of a topic node, wherein the topic node is a node in an information tree representing a product information system, and the question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the degree of contribution of the question to product sales;

[0024] Determine a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, and adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists;

[0025] For each first question list, display the first question list through the first terminal, and receive first operation information of the user for the first question list through the first terminal;

[0026] Calculate the click value score of each question in the question contribution queue according to each first operation information;

[0027] Inputting the problem sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value of the problem sorting exchange combination;

[0028] Calculating a model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain an exchange prediction model;

[0029] Inputting the problem sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value;

[0030] The first question list is adjusted according to the obtained exchange evaluation value to obtain a second question list, and the second question list is used to be displayed through the second terminal to achieve information recommendation.

[0031] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0032] Obtaining a question contribution queue of a topic node, wherein the topic node is a node in an information tree representing a product information system, and the question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the degree of contribution of the question to product sales;

[0033] Determine a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, and adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists;

[0034] For each first question list, display the first question list through the first terminal, and receive first operation information of the user for the first question list through the first terminal;

[0035] Calculate the click value score of each question in the question contribution queue according to each first operation information;

[0036] Inputting the problem sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value of the problem sorting exchange combination;

[0037] Calculating a model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain an exchange prediction model;

[0038] Inputting the problem sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value;

[0039] The first question list is adjusted according to the obtained exchange evaluation value to obtain a second question list, and the second question list is used to be displayed through the second terminal to achieve information recommendation.

[0040] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects: obtaining a question contribution queue, and arranging the questions in the queue in descending order according to their contribution to product sales; determining a question sorting exchange combination of the question contribution queue to adjust the question contribution queue to obtain several first question lists; calculating the click value score of each question based on the user's operation information on the first question list, and the click value score measures the click value of the question and the user's interest level; inputting the question sorting exchange combination into the initial exchange prediction model to obtain an exchange prediction evaluation value, and the exchange prediction evaluation value measures the probability of exchanging the question sort according to a certain question sorting exchange combination; calculating the model loss based on the exchange prediction evaluation value and the click value score, which can take into account the click value of the question and the probability of exchanging the question sort; and performing reinforcement learning based on the model loss can obtain the exchange prediction model; the exchange prediction model can output the exchange evaluation value of the question sorting exchange combination, and select the question sorting exchange combination according to the exchange evaluation value to perform question sorting exchange, so that the newly generated second question list takes into account the contribution of the question and the user's click probability at the same time, and the user can obtain useful product information faster based on the second question list, thereby improving the efficiency of information recommendation through the second question list. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0043] Figure 2 is a flow chart of an embodiment of the information recommendation method according to the present application;

[0044] Figure 3 is a flowchart of generating a contribution queue in one embodiment of the present application;

[0045] Figure 4 is a schematic diagram of an information tree in one embodiment of the present application;

[0046] Figure 5 is a schematic diagram of a value problem chain in one embodiment of the present application;

[0047] Figure 6 is a structural schematic diagram of an embodiment of an information recommendation device according to the present application;

[0048] Figure 7 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0050] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0052] like Figure 1As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0053] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0054] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, etc.

[0055] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0056] It should be noted that the information recommendation method provided in the embodiment of the present application is generally executed by a server, and accordingly, the information recommendation device is generally arranged in the server.

[0057] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0058] Continue to refer Figure 2 , shows a flow chart of an embodiment of an information recommendation method according to the present application. The information recommendation method comprises the following steps:

[0059] Step S201, obtaining a question contribution queue of a topic node, where a topic node is a node in an information tree representing a product information system. The question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the degree of contribution of a question to product sales.

[0060] In this embodiment, the information recommendation method is executed on the electronic device (eg Figure 1 The server shown in the figure) can communicate with the terminal through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultrawideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0061] The information tree reflects the product information system in the form of a tree, and the product information system covers the conversation scenarios and topics that may be involved in the product consultation and purchase scenario. There are topic nodes in the information tree, and each topic node corresponds to a scenario during consultation and purchase. For example, product definition can be a topic node.

[0062] Specifically, each topic node corresponds to a question contribution queue, which includes several questions. The questions can be sorted in descending order according to the value of contribution. The contribution measures the contribution of the question to the product sold. The question contribution queue needs to be calculated in advance.

[0063] Step S202: determining a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, and adjusting the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists.

[0064] Specifically, an exchange strategy is preset in the server, and the questions in the question contribution queue are combined according to the exchange strategy to obtain a question sorting exchange combination. The question sorting exchange combination contains two questions, and the server exchanges the sorting positions of the two questions in the question contribution queue to obtain a first question list corresponding to the question sorting exchange combination. According to the exchange strategy, multiple question sorting exchange combinations can be obtained, each of which corresponds to a first question list, and all question sorting exchange combinations can exchange the sorting positions of all questions in the question contribution queue at least once.

[0065] In one embodiment, the exchange strategy is: select the two questions with the highest and lowest contribution, the two questions with the second highest and second lowest contribution from the question contribution queue in turn, and so on and so forth in this order. If the number of questions in the question contribution queue is an odd number, finally randomly select a question and combine it with the question with the middle contribution to obtain a question sorting exchange combination.

[0066] Step S203: for each first question list, display the first question list through the first terminal, and receive first operation information of the user for the first question list through the first terminal.

[0067] Specifically, the first question list can be displayed through the first terminal, and the first terminal can be a terminal operated by the user. The user can click on the questions in the first question list through the first terminal to obtain the product information corresponding to the question; in addition, for the displayed questions, the user can not click, or give feedback that the question quality is poor. The first terminal records the user's click behavior, non-click behavior, and feedback behavior on the question, obtains the first operation information, and sends the first operation information to the server.

[0068] Since there are multiple first question lists, after detecting each click or feedback behavior of the user, the next first question list is displayed. For example, the first question list displayed for the first time is obtained by exchanging the two questions with the highest and lowest contribution, and the first question list displayed for the second time is obtained by exchanging the two questions with the second highest and second lowest contribution.

[0069] Step S204, calculating the click value score of each question in the question contribution queue according to each piece of first operation information.

[0070] Specifically, according to the record of the first operation information, different behaviors correspond to different scores. Positive scores can be assigned to questions corresponding to click behaviors, zero scores can be assigned to questions without click behaviors, and negative scores can be assigned to questions corresponding to feedback behaviors. Feedback behaviors can be users complaining about the poor quality of questions.

[0071] In one embodiment, the sorting positions of two questions in the question sorting exchange combination are exchanged to obtain a first question list; when the first question list is displayed through the first terminal, if a click behavior occurs on one of the two questions, a positive score is assigned to the question where the click behavior occurs, for example, +1 point; if one of the two questions is not clicked, a zero point is assigned to the question where the click behavior does not occur; if feedback behavior occurs on one of the two questions, a negative score is assigned to the question where feedback behavior occurs, for example, -1 point.

[0072] The question sorting exchange combination corresponds to the first question list, and the first question list corresponds to the first operation information. According to the score of each question in each first operation information, the click value score of each question can be calculated. For a question, its click value score is calculated by the statistics of its corresponding scores, which can reflect the click value of the question and the degree of user interest in the question.

[0073] Furthermore, the above-mentioned step S204 may include: obtaining the direct value score of each question in the question contribution queue according to each first operation information; calculating the indirect value score of each question based on the direct value score and a preset attenuation coefficient; and accumulating the direct value score and indirect value score of each question respectively to obtain the click value score of each question.

[0074] Specifically, based on the first operation information, the score obtained by the question in each round of display can be obtained, and this score is the direct value score of the question. After each round of display, the score of the question will have an impact on subsequent user actions, and this impact will decay as the number of rounds increases. The impact of the question on subsequent actions can be measured by the indirect value score, which is obtained by multiplying the direct value score by the power of the attenuation coefficient, where the exponent of the attenuation coefficient power is the number of rounds of the interval. For example, the direct value score of question A in the i-th round is S, and the attenuation coefficient is γ. The indirect value score generated by the direct value score S in the i+2 round is γ 2 ·S.

[0075] The direct value scores of the questions are accumulated, and the indirect value scores are accumulated, and then the accumulated sum of the direct value scores and the accumulated sum of the indirect value scores are added to obtain the click value score of the question.

[0076] Assume that the score obtained for the i-th round of questions is Define the decay coefficient of the score as γ, then the final click value score is:

[0077]

[0078] in, is the cumulative sum of the direct value scores of the question, It is the cumulative sum of the indirect value scores of the question.

[0079] In this embodiment, the score obtained by the question in each round of question dialogue will also affect the subsequent question dialogue. Therefore, in addition to calculating the direct value score of the question, the indirect value score of the question is also calculated to ensure the accuracy of the final click value score.

[0080] Step S205, input the question sorting exchange combination into the initial exchange prediction model to obtain the exchange prediction evaluation value of the question sorting exchange combination.

[0081] The initial exchange prediction model may be an exchange prediction model that has not yet been trained, and the exchange prediction model is used to predict whether to exchange the sorting positions of the questions in the question sorting exchange combination.

[0082] Specifically, the questions in the question sorting exchange combination are converted into vectors, and then input into the pre-built initial exchange prediction model. The initial exchange prediction model is constructed according to the exchange strategy. If N (N is a positive integer greater than 1) groups of question sorting exchange combinations are obtained according to the exchange strategy, then the initial exchange prediction model is a multi-layer neural network for N actions. Each question sorting exchange combination corresponds to an action. The neural network needs to predict the N actions and output the probability of executing each action, that is, adjust the probability of the first question list according to the question action combination.

[0083] The model parameters of the initial exchange prediction model can be randomly initialized using the Xavier method, which is a very effective neural network initialization method.

[0084] The initial exchange prediction model may include several hidden layers and an output layer. In one embodiment, the initial exchange prediction model includes two hidden layers. Suppose the weight matrix of the first hidden layer is w1, its activation function is the linear rectified function (Rectified Linear Unit, ReLU), and the bias is b1. Then the output of the first hidden layer is o1=relu(w1*v+b1), where v is the input vector; suppose the weight matrix of the second hidden layer is w2, the bias is b2, and the activation function is the ReLU function, then the output of the second hidden layer is: o2=relu(w2*o1+b2). ​​The output layer can use a softmax layer, which can implement a logistic regression algorithm, and its output o3 is the exchange prediction evaluation value Pa. In order to improve the prediction effect, the number of hidden layers can be increased.

[0085] Step S206, calculating the model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss and obtain the exchange prediction model.

[0086] Specifically, the click value score measures the click value of the question and the degree of user interest in the question. The more click-worthy and interesting the question is, the more important it is when swapping ranking positions. It needs to be kept as high as possible so that users can find and click on these questions in time. The swap prediction evaluation value is the model's prediction of whether to swap the ranking position of the question.

[0087] Therefore, the present application adopts a reinforcement learning training method so that the model can output the prediction results of sorting exchange questions in accordance with the obtained click value scores. During the training, based on the preset loss function calculation method, the obtained exchange prediction evaluation value and click value score are calculated to obtain the model loss. The model parameters of the initial exchange prediction model are adjusted according to the model loss, and the initial exchange prediction model after the adjusted parameters is iteratively trained until the training is completed to obtain the exchange prediction model.

[0088] Furthermore, the above step S204 may include: calculating the model loss based on the obtained exchange prediction evaluation value and click value score; adjusting the initial exchange prediction model and continuing training with maximizing the model loss as the training goal, and stopping the training until the obtained model loss converges to obtain the exchange prediction model.

[0089] Specifically, each question corresponds to a click value score and an exchange prediction evaluation value. The negative value of the product of the click value score and the exchange prediction evaluation value can be used as the model loss. The server takes maximizing the model loss as the training goal, adjusts the model parameters in the initial exchange prediction model, and continues to train the model after adjusting the parameters until the obtained model loss reaches convergence, stops training, and obtains the exchange prediction model.

[0090] The server can use SGD (stochastic gradient descent) and Adam algorithm (an algorithm that optimizes SGD and can more effectively update network weights) to adjust the parameters of the initial exchange prediction model.

[0091] In this embodiment, after calculating the model loss based on the exchange prediction evaluation value and the click value score, the initial exchange prediction model is optimized with maximizing the model loss as the training goal to obtain the exchange prediction model. The exchange prediction model can maximize the click value score of each question and can better sort the questions.

[0092] Step S207, input the question sorting exchange combination into the exchange prediction model to obtain the exchange evaluation value.

[0093] Specifically, after obtaining the exchange prediction model, the exchange prediction model can be used to recommend product information. The server inputs the question ranking exchange combination into the exchange prediction model, and the exchange prediction model can output an exchange evaluation value for each question ranking exchange combination. For a question ranking exchange combination, the higher the corresponding exchange evaluation value, the higher the probability that the model will exchange the ranking positions of the two questions in the question ranking combination.

[0094] Step S208: adjusting the first question list according to the obtained exchange evaluation value to obtain a second question list, where the second question list is used for display through the second terminal to achieve information recommendation.

[0095] Specifically, the server selects the exchange evaluation value with the largest value from the obtained exchange evaluation values, and exchanges the sorting positions of the two questions corresponding to the exchange evaluation value in the first question list to obtain the second question list.

[0096] The second question list can be displayed through the second terminal, and when a question in the second question list is clicked, the product information corresponding to the question is displayed through the second terminal. The exchange prediction model can determine how to exchange the order of the questions in the first question list, and put the questions that the user may be most interested in as far as possible in the front position, thereby increasing the user's click probability. The user can obtain the product information of interest more quickly through the second question list, which improves the efficiency of information recommendation through the second question list.

[0097] In one embodiment, when it is detected that a user clicks on a question in the second question list, the question sorting exchange combination is regenerated, the exchange evaluation value is recalculated by the exchange prediction model, and then the exchange evaluation value is screened to update the second question list to obtain a new second question list. The newly generated second question list is then displayed by the second terminal, thereby realizing real-time update of the second question list during user access.

[0098] In this embodiment, a question contribution queue is obtained, and the questions in the queue are arranged in descending order according to their contribution to product sales; a question sorting exchange combination of the question contribution queue is determined to adjust the question contribution queue to obtain a plurality of first question lists; the click value score of each question is calculated based on the user's operation information on the first question list, and the click value score measures the click value of the question and the user's interest level; the question sorting exchange combination is input into the initial exchange prediction model to obtain an exchange prediction evaluation value, and the exchange prediction evaluation value measures the probability of exchanging the question sort according to a certain question sorting exchange combination; the model loss is calculated based on the exchange prediction evaluation value and the click value score, which can take into account the click value of the question and the probability of exchanging the question sorting, and reinforcement learning can be performed based on the model loss to obtain the exchange prediction model; the exchange prediction model can output the exchange evaluation value of the question sorting exchange combination, and the question sorting exchange combination is selected according to the exchange evaluation value to perform the sorting exchange of the question, so that the newly generated second question list takes into account the contribution of the question and the user's click probability at the same time, and the user can obtain useful product information more quickly based on the second question list, thereby improving the efficiency of information recommendation through the second question list.

[0099] Further, such as Figure 3 As shown, in one embodiment, before the above step S201, a step of generating a contribution queue is also included, and the step may include:

[0100] Step S209, obtaining an information tree representing the product information system and questions related to the product information, wherein the information tree includes a scene node, and the child nodes of the scene node are topic nodes.

[0101] Specifically, the information tree reflects the product information system in the form of a tree, and the product information system covers the conversation scenarios and topics that may be involved in the product consultation and purchase scenarios. The information tree contains scenario nodes and topic nodes. The granularity of scenario nodes and topic nodes is different. The granularity and scope of topic nodes are smaller than those of scenario nodes. They are a further division and confirmation of customer intentions. Topic nodes are child nodes of scenario nodes.

[0102] Figure 4 This is a schematic diagram of an information tree for completing problem location in an embodiment, where the root node of the information tree is the product, and the scenario nodes in the conversation may include: product consultation, process consultation and after-sales consultation; wherein the topic nodes included in product consultation are: product features, product prices and logistics links; wherein the issues included in the product features scenario node include: how big is the product, what colors are available for the product, what functions does the product have, etc.

[0103] Specifically, the server may obtain an information tree and a plurality of questions. When these questions are displayed to the user, the user may click on these questions to obtain relevant product information, and different questions are associated with different product information.

[0104] Step S210, determining the topic node corresponding to the question according to the scene topic attribution of the question, so as to obtain a third question list of the topic node.

[0105] Specifically, the server needs to determine the topic node corresponding to each question, so as to locate the question in the information tree. The semantic similarity between the question and each topic node can be calculated, and the semantic similarity can be used as the scene topic attribution degree of the question under the scene node. The topic node corresponding to the maximum scene topic attribution degree is selected as the topic node to which the question belongs, so as to locate each question in the information tree.

[0106] For each topic node, the questions included in the topic node constitute the third question list of the topic node.

[0107] Furthermore, the above step S210 may include: calculating the scene attribution of each scene node based on semantics; calculating the topic attribution of each topic node based on semantics; calculating the scene-topic attribution of the question to each topic node based on the scene attribution and the topic attribution; selecting the topic node corresponding to the maximum scene-topic attribution from each scene-topic attribution as the topic node corresponding to the question; and arranging the questions corresponding to the topic nodes in descending order according to the scene-topic attribution to obtain a third question list of the topic nodes.

[0108] Specifically, the server calculates the scene attribution degree between the question and each scene node from a semantic level, and the scene attribution degree represents the relevance between the question and the scene node; it also needs to calculate the topic attribution degree between the question and each topic node from a semantic level, and the topic attribution degree represents the relevance between the question and the topic node.

[0109] In one embodiment, the server inputs the question into a trained classification model, which may be a Bert model. The classification model converts the question Q i Convert it into a vector representation, and then calculate the semantic similarity with each scene node to obtain the scene attribution Calculate the semantic similarity between the vector representation and each topic node to obtain the topic attribution

[0110] Scenario attribution of question and topic nodes By scene attribution and topic attribution Multiplying together gives:

[0111]

[0112] Among them, the scene node corresponding to the scene attribution degree used in the calculation should be the parent node of the topic node corresponding to the topic attribution degree used.

[0113] The server selects the topic node corresponding to the maximum scene topic attribution degree as the topic node to which the question belongs, thereby locating the question in the information tree.

[0114] For each topic node, the questions included in the topic node are sorted in descending order according to the scene topic attribution degree to obtain the third question list of the topic node.

[0115] In this embodiment, the scene attribution and topic attribution of the question are calculated first, and then the scene topic attribution of the question is calculated, so as to locate the question to the topic node in the information tree, and the questions contained in the topic node can be sorted according to the scene topic attribution to obtain the third question list of each topic node.

[0116] Step S211, obtaining the third operation information of the user on the third question list, so as to determine a valuable question chain according to the third operation information.

[0117] Specifically, the user can browse the product through the terminal, the server can display the third question list through the terminal, and the user clicks on the questions in the third question list to obtain the product information and choose to purchase or not purchase the product.

[0118] The server can display multiple third question lists, and different third question lists display questions under different scenarios and topics. The user's clicks and browsing behaviors on the third question list, as well as whether the user purchases the product, will all be recorded as third operation information.

[0119] According to the third operation information, all the questions that the user clicks and browses in the process from logging in to completing a product purchase can be obtained. These questions constitute a valuable question chain.

[0120] When the terminal displays the third question list, the questions in the list are displayed in descending order of scene topic attribution. When the display page is limited, K (K is less than the total number of questions in the third question list) questions can be displayed in the display area (for example, K can be 3, 4, or 5), and users can slide questions to browse other questions.

[0121] The terminal can display scene nodes and topic nodes in the form of labels in the display area. When the label of the topic node is clicked, the third question list corresponding to the topic node is displayed. When a question is clicked, the answer to the question is displayed, and other unaccessed questions are displayed below the answer.

[0122] In one embodiment, the user can enter the question Q through the terminal m , the server calculates the user input question Q m Similarity with existing questions, determine the most similar question Q n , the most similar question Q n The answers to the most similar questions are displayed first. n The third question list of the corresponding scene node.

[0123] Furthermore, the above-mentioned step S211 may include: displaying a third question list through a third terminal, and receiving third operation information of the user for the third question list through the third terminal, the third operation information including several click question chains, each click question chain including product price and question click sequence information; for each click question chain, obtaining the purchase result of the click question chain; when it is determined that the product purchase is successful based on the purchase result, determining the click question chain as a valuable question chain.

[0124] Specifically, the server can display a third question list through a third terminal. A user can click on a question in the third question list. The clicked questions can form a click question chain. The clicked questions are questions related to a certain product. The product has a product price, so the click question chain also includes the product price; the questions clicked by the user have a sequence, and the sequence of the questions is recorded as question click sequence information, so the click question chain also has question click sequence information; the click question chain also has a purchase result, which records whether the user has purchased the product pointed to by the click question chain.

[0125] The third terminal can record the click question chain, product price, question click sequence information and purchase results to obtain third operation information; the third operation information can include multiple click question chains.

[0126] The server obtains the third operation information from the third terminal, extracts the purchase result of the click question chain therefrom, and if the purchase result indicates that the product is purchased by the user, the click question chain is determined as a valuable question chain. Figure 5 is a schematic diagram of a value question chain in an embodiment, where a user clicks on three questions in sequence and then completes the purchase of a product.

[0127] In this embodiment, the click question chain in the third operation information has a purchase result, and the purchase result records whether the user purchases the product. The click question chain that generates a purchase behavior is more important for information recommendation and can be marked as a valuable question chain.

[0128] Step S212: for each user community, according to the valuable question chain of each user in the user community, calculate the contribution of each question in the valuable question chain.

[0129] Specifically, products can be clustered according to their characteristics, thereby dividing products into different categories. Users who have purchased the same category of products will be divided into the same user community.

[0130] The contribution reflects the degree of contribution of the question to the product sales. The contribution needs to be calculated based on the valuable question chain. The higher the product price corresponding to the valuable question chain, the fewer the number of questions, and the greater the contribution of each question to the product. In one embodiment, the product price is evenly divided according to the number of questions in the valuable question chain to obtain the contribution of each question in the valuable question chain.

[0131] The contribution degree in this application is calculated based on the user community level. There are multiple users in a user community, and each user can have multiple valuable question chains. The contribution degree of each valuable question chain can be calculated. The user community is associated with multiple questions. For each question, the contribution degree of the question in each valuable question chain is accumulated to obtain the contribution degree of the question. This contribution degree measures the contribution degree of the question within the user community.

[0132] Furthermore, the above step S212 may include: for each user community, obtaining the valuable question chain of each user in the user community; for each obtained valuable question chain, calculating the single-chain contribution of each question in the valuable question chain according to the product price and question click order information in the valuable question chain; within the user community, counting the contribution of each question in the valuable question chain of each user according to the calculated single-chain contribution.

[0133] Specifically, for each valuable question chain involved in the user community, obtain the product price sale_sum of its related products and the question click order information. Assume that there is a question Q in the valuable question chain i , Question Q i The single chain contribution is C i , then:

[0134]

[0135]

[0136] in, is the contribution coefficient, m is the number of questions in the valuable question chain, k is the ordinal number of the question being clicked, which is obtained from the question click order information. The representative question is the kth question among the m questions in the valuable question chain and is also the kth question clicked by the user.

[0137] According to all the valuable question chains in the user community, the contribution of each single chain of the question is added up to get the contribution of the question, which is the total contribution degree of the question within the user community.

[0138] In one embodiment, after each cycle, the server re-determines the topic node corresponding to the question and updates the third question list of the topic node; obtains the user's third operation information for the updated third topic list, so as to recalculate the contribution of each question based on the valuable question chain in the new third operation information.

[0139] In this embodiment, the user community contains multiple valuable question chains. The single-chain contribution of a question in a valuable question chain is calculated first, and then the total contribution of the question is calculated. The contribution is calculated from the user community level, taking into account the common attributes between users, and can more accurately measure the contribution of the question to product recommendations.

[0140] Step S213: for each topic node, a contribution queue of the topic node is generated according to the contribution of each question corresponding to the topic node.

[0141] Specifically, there are multiple topic nodes in the information tree, and for each topic node, the contribution of the questions associated with it has been obtained. According to the order of contribution from large to small, the questions related to the topic node are sorted to obtain the contribution queue of the topic node.

[0142] In this embodiment, the position of the question in the information tree is determined to obtain the third question list of the topic node; the third question list is displayed to obtain the user's third operation information, and a valuable question chain is obtained based on the third operation information. The valuable question chain records the price of successfully purchased products and the order in which the user's question clicks, so that the contribution of the question can be calculated and the contribution queue of each topic node can be obtained based on the contribution. The generation of the contribution queue takes into account the user's community attributes and more accurately reflects the contribution of the question to product recommendations and the user's interest in the question.

[0143] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0144] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0145] Further references Figure 6 , as a response to the above Figure 2 The present application provides an embodiment of an information recommendation device, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0146] like Figure 6 As shown, the information recommendation device 300 described in this embodiment includes: a queue acquisition module 301, a sorting adjustment module 302, an information receiving module 303, a score calculation module 304, a combination input module 305, a model adjustment module 306, a question input module 307 and a list adjustment module 308, wherein:

[0147] The queue acquisition module 301 is used to obtain the question contribution queue of the topic node. The topic node is a node in the information tree representing the product information system. The question contribution queue includes a number of questions arranged according to the contribution. The contribution is used to measure the contribution of the question to the product sales.

[0148] The sorting adjustment module 302 is used to determine the question sorting exchange combination of the question contribution queue based on a preset exchange strategy, so as to adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists.

[0149] The information receiving module 303 is used to display the first question list through the first terminal for each first question list, and receive the first operation information of the user for the first question list through the first terminal.

[0150] The score calculation module 304 is used to calculate the click value score of each question in the question contribution queue according to each first operation information.

[0151] The combination input module 305 is used to input the question sorting exchange combination into the initial exchange prediction model to obtain the exchange prediction evaluation value of the question sorting exchange combination.

[0152] The model adjustment module 306 is used to calculate the model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain the exchange prediction model.

[0153] The question input module 307 is used to input the question sorting exchange combination into the exchange prediction model to obtain the exchange evaluation value.

[0154] The list adjustment module 308 is used to adjust the first question list according to the obtained exchange evaluation value to obtain a second question list, and the second question list is used to be displayed through the second terminal to achieve information recommendation.

[0155] In this embodiment, a question contribution queue is obtained, and the questions in the queue are arranged in descending order according to their contribution to product sales; a question sorting exchange combination of the question contribution queue is determined to adjust the question contribution queue to obtain a plurality of first question lists; the click value score of each question is calculated based on the user's operation information on the first question list, and the click value score measures the click value of the question and the user's interest level; the question sorting exchange combination is input into the initial exchange prediction model to obtain an exchange prediction evaluation value, and the exchange prediction evaluation value measures the probability of exchanging the question sort according to a certain question sorting exchange combination; the model loss is calculated based on the exchange prediction evaluation value and the click value score, which can take into account the click value of the question and the probability of exchanging the question sorting, and reinforcement learning can be performed based on the model loss to obtain the exchange prediction model; the exchange prediction model can output the exchange evaluation value of the question sorting exchange combination, and the question sorting exchange combination is selected according to the exchange evaluation value to perform the sorting exchange of the question, so that the newly generated second question list takes into account the contribution of the question and the user's click probability at the same time, and the user can obtain useful product information more quickly based on the second question list, thereby improving the efficiency of information recommendation through the second question list.

[0156] In some optional implementations of this embodiment, the information recommendation device 300 may further include: an acquisition module, a node determination module, an information acquisition module, a contribution calculation module, and a queue generation module, wherein:

[0157] The acquisition module is used to acquire an information tree representing a product information system and questions related to the product information. The information tree includes a scene node, and the child nodes of the scene node are topic nodes.

[0158] The node determination module is used to determine the topic node corresponding to the question according to the scene topic attribution of the question, so as to obtain a third question list of the topic node.

[0159] The information acquisition module is used to acquire the third operation information of the user for the third question list, so as to determine the valuable question chain according to the third operation information.

[0160] The contribution calculation module is used to calculate the contribution of each question in the valuable question chain for each user community according to the valuable question chain of each user in the user community.

[0161] The queue generation module is used to generate a contribution queue of each topic node according to the contribution of each question corresponding to the topic node.

[0162] In this embodiment, the position of the question in the information tree is determined to obtain the third question list of the topic node; the third question list is displayed to obtain the user's third operation information, and a valuable question chain is obtained based on the third operation information. The valuable question chain records the price of successfully purchased products and the order in which the user's question clicks, so that the contribution of the question can be calculated and the contribution queue of each topic node can be obtained based on the contribution. The generation of the contribution queue takes into account the user's community attributes and more accurately reflects the contribution of the question to product recommendations and the user's interest in the question.

[0163] In some optional implementations of this embodiment, the node determination module may include: a scene calculation submodule, a topic calculation submodule, an attribution calculation submodule, a node selection submodule, and a question arrangement submodule, wherein:

[0164] The scene calculation submodule is used to calculate the scene attribution of each scene node based on semantic calculation problems.

[0165] The topic calculation submodule is used to calculate the topic attribution of each topic node based on semantic calculation problems.

[0166] The attribution calculation submodule is used to calculate the scene topic attribution degree of the question to each topic node based on the scene attribution degree and the topic attribution degree.

[0167] The node selection submodule is used to select the topic node corresponding to the maximum scene topic attribution degree from the scene topic attribution degrees as the topic node corresponding to the question.

[0168] The question arrangement submodule is used to arrange the questions corresponding to the topic nodes in descending order according to the scene topic attribution degree, and obtain the third question list of the topic node.

[0169] In this embodiment, the scene attribution and topic attribution of the question are calculated first, and then the scene topic attribution of the question is calculated, so as to locate the question to the topic node in the information tree, and the questions contained in the topic node can be sorted according to the scene topic attribution to obtain the third question list of each topic node.

[0170] In some optional implementations of this embodiment, the information acquisition module may include: an information receiving submodule, a result acquisition submodule, and a determination submodule, wherein:

[0171] The information receiving submodule is used to display the third question list through the third terminal, and receive the user's third operation information on the third question list through the third terminal. The third operation information includes several click question chains, and each click question chain includes product price and question click order information.

[0172] The result acquisition submodule is used to acquire the purchase result of each click question chain.

[0173] The determination submodule is used to determine the click question chain as a valuable question chain when determining that the product purchase is successful based on the purchase result.

[0174] In this embodiment, the click question chain in the third operation information has a purchase result, and the purchase result records whether the user purchases the product. The click question chain that generates a purchase behavior is more important for information recommendation and can be marked as a valuable question chain.

[0175] In some optional implementations of this embodiment, the contribution calculation module may include: a question chain acquisition submodule, a single chain calculation submodule and a contribution statistics submodule, wherein:

[0176] The question chain acquisition submodule is used to obtain the valuable question chains of each user in each user community.

[0177] The single-chain calculation submodule is used to calculate the single-chain contribution of each question in the valuable question chain according to the product price and question click sequence information in the valuable question chain for each acquired valuable question chain.

[0178] The contribution statistics submodule is used to count the contribution of each question in each user's valuable question chain within the user community based on the calculated single-chain contribution.

[0179] In this embodiment, the user community contains multiple valuable question chains. The single-chain contribution of a question in a valuable question chain is calculated first, and then the total contribution of the question is calculated. The contribution is calculated from the user community level, taking into account the common attributes between users, and can more accurately measure the contribution of the question to product recommendations.

[0180] In some optional implementations of this embodiment, the score calculation module 304 may include: a direct acquisition submodule, an indirect calculation submodule, and a score accumulation submodule, wherein:

[0181] The direct acquisition submodule is used to obtain the direct value score of each question in the question contribution queue according to each first operation information.

[0182] The indirect calculation submodule is used to calculate the indirect value score of each question based on the direct value score and a preset attenuation coefficient.

[0183] The score accumulation submodule is used to accumulate the direct value score and the indirect value score of each question respectively to obtain the click value score of each question.

[0184] In this embodiment, the score obtained by the question in each round of question dialogue will also affect the subsequent question dialogue. Therefore, in addition to calculating the direct value score of the question, the indirect value score of the question is also calculated to ensure the accuracy of the final click value score.

[0185] In some optional implementations of this embodiment, the model adjustment module 306 may include: a loss calculation submodule and a model adjustment submodule, wherein:

[0186] The loss calculation submodule is used to calculate the model loss based on the obtained exchange prediction evaluation value and click value score.

[0187] The model adjustment submodule is used to adjust the initial exchange prediction model and continue training with maximizing the model loss as the training goal, and stop training when the obtained model loss converges to obtain the exchange prediction model.

[0188] In this embodiment, after calculating the model loss based on the exchange prediction evaluation value and the click value score, the initial exchange prediction model is optimized with maximizing the model loss as the training goal to obtain the exchange prediction model. The exchange prediction model can maximize the click value score of each question and can better sort the questions.

[0189] To solve the above technical problems, the present application also provides a computer device. Figure 7 , Figure 7 This is a basic structural block diagram of the computer device in this embodiment.

[0190] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0191] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0192] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the information recommendation method, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0193] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the information recommendation method.

[0194] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0195] The computer device provided in this embodiment can execute the above information recommendation method. Here, the information recommendation method can be the information recommendation method of each of the above embodiments.

[0196] In this embodiment, the question contribution queue is adjusted according to the exchange evaluation value output by the exchange prediction model to obtain a second question list. The second question list takes into account both the contribution of the question to product sales and the user's click probability. Based on the second question list, the user can obtain useful product information more quickly, thereby improving the efficiency of information recommendation through the second question list.

[0197] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the information recommendation method as described above.

[0198] In this embodiment, the question contribution queue is adjusted according to the exchange evaluation value output by the exchange prediction model to obtain a second question list. The second question list takes into account both the contribution of the question to product sales and the user's click probability. Based on the second question list, the user can obtain useful product information more quickly, thereby improving the efficiency of information recommendation through the second question list.

[0199] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0200] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An information recommendation method, characterized in that: The steps include: Obtaining a question contribution queue of a topic node, wherein the topic node is a node in an information tree representing a product information system, and the question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the degree of contribution of the question to product sales; Determine a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, and adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists; For each first question list, display the first question list through the first terminal, and receive first operation information of the user for the first question list through the first terminal; Calculate the click value score of each question in the question contribution queue according to each first operation information; Inputting the problem sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value of the problem sorting exchange combination; Calculating a model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain an exchange prediction model; Inputting the problem sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value; The first question list is adjusted according to the obtained exchange evaluation value to obtain a second question list, wherein the second question list is used for displaying through a second terminal to achieve information recommendation; Before the step of obtaining the question contribution queue of the topic node, the method further includes: Acquire an information tree representing a product information system and questions related to the product information, wherein the information tree includes a scene node, and a child node of the scene node is a topic node; Determine the topic node corresponding to the question according to the scene topic attribution of the question, so as to obtain a third question list of the topic node; Acquiring third operation information of the user on the third question list, so as to determine a valuable question chain according to the third operation information; For each user community, according to the valuable question chain of each user in the user community, the contribution of each question in the valuable question chain is calculated; For each topic node, a contribution queue of the topic node is generated according to the contribution of each question corresponding to the topic node.

2. The information recommendation method according to claim 1, characterized in that: The step of determining the topic node corresponding to the question according to the scene topic attribution of the question to obtain a third question list of the topic node includes: Calculate the scene belongingness of the problem to each scene node based on semantics; Calculate the topic attribution of the question to each topic node based on semantics; Calculating the scene topic attribution degree of the question to each topic node according to the scene attribution degree and the topic attribution degree; From the topic attribution degrees of each scene, select the topic node corresponding to the maximum scene topic attribution degree as the topic node corresponding to the question; According to the scene topic attribution degree, the questions corresponding to the topic node are arranged in descending order to obtain a third question list of the topic node.

3. The information recommendation method according to claim 1, characterized in that: The step of obtaining the third operation information of the user on the third question list to determine the valuable question chain according to the third operation information includes: Displaying the third question list through a third terminal, and receiving third operation information of the user for the third question list through the third terminal, wherein the third operation information includes a plurality of click question chains, each click question chain includes product price and question click sequence information; For each click question chain, obtain the purchase results of the click question chain; When it is determined that the product purchase is successful according to the purchase result, the click question chain is determined as a valuable question chain.

4. The information recommendation method according to claim 3, characterized in that: The step of calculating the contribution of each question in the valuable question chain for each user community according to the valuable question chain of each user in the user community includes: For each user community, obtain the valuable question chain of each user in the user community; For each valuable question chain obtained, the single-chain contribution of each question in the valuable question chain is calculated according to the product price and question click order information in the valuable question chain; In the user community, the contribution of each question in each user's valuable question chain is counted based on the calculated single-chain contribution.

5. The information recommendation method according to claim 1, characterized in that: The step of calculating the click value score of each question in the question contribution queue according to each first operation information includes: According to each piece of first operation information, a direct value score of each question in the question contribution queue is obtained; Calculate the indirect value score of each of the questions based on the direct value score and a preset attenuation coefficient; The direct value score and the indirect value score of each question are respectively accumulated to obtain the click value score of each question.

6. The information recommendation method according to claim 1, characterized in that: The step of calculating the model loss according to the obtained exchange prediction evaluation value and the click value score to adjust the initial exchange prediction model according to the model loss to obtain the exchange prediction model includes: Calculate the model loss based on the obtained exchange prediction evaluation value and click value score; Taking maximizing the model loss as the training goal, the initial exchange prediction model is adjusted and training is continued until the obtained model loss converges, and the training is stopped to obtain the exchange prediction model.

7. An information recommendation device, characterized in that: include: A queue acquisition module is used to acquire a question contribution queue of a topic node, wherein the topic node is a node in an information tree representing a product information system, and the question contribution queue includes a number of questions arranged according to contribution, and the contribution is used to measure the contribution of the question to product sales; A sorting adjustment module, used for determining a question sorting exchange combination of the question contribution queue based on a preset exchange strategy, so as to adjust the sorting of the question contribution queue according to the question sorting exchange combination to obtain a plurality of first question lists; An information receiving module, configured to display, for each first question list, the first question list through a first terminal, and receive, through the first terminal, first operation information of a user with respect to the first question list; A score calculation module, used to calculate the click value score of each question in the question contribution queue according to each first operation information; A combination input module, used for inputting the problem sorting exchange combination into an initial exchange prediction model to obtain an exchange prediction evaluation value of the problem sorting exchange combination; A model adjustment module, used for calculating a model loss according to the obtained exchange prediction evaluation value and click value score, so as to adjust the initial exchange prediction model according to the model loss to obtain an exchange prediction model; A question input module, used for inputting the question sorting exchange combination into the exchange prediction model to obtain an exchange evaluation value; A list adjustment module, used for adjusting the first question list according to the obtained exchange evaluation value to obtain a second question list, wherein the second question list is used for displaying through a second terminal to realize information recommendation; Before obtaining the question contribution queue of the topic node, the method further includes: Acquire an information tree representing a product information system and questions related to the product information, wherein the information tree includes a scene node, and a child node of the scene node is a topic node; Determine the topic node corresponding to the question according to the scene topic attribution of the question, so as to obtain a third question list of the topic node; Acquiring third operation information of the user on the third question list, so as to determine a valuable question chain according to the third operation information; For each user community, according to the valuable question chain of each user in the user community, the contribution of each question in the valuable question chain is calculated; For each topic node, a contribution queue of the topic node is generated according to the contribution of each question corresponding to the topic node.

8. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the information recommendation method according to any one of claims 1 to 6 when executing the computer-readable instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the information recommendation method according to any one of claims 1 to 6 are implemented.

Images