Information pushing method and device, electronic equipment and medium
By calculating the comprehensive score of target product information, the problem of new product push strategy for e-commerce platforms is solved, precise push and operational effect are improved, and costs are reduced.
Patent Information
- Application Number
- CN202310540477.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-12
- Publication Date
- 2024-11-12
AI Technical Summary
On the e-commerce platform, how to determine the appropriate operational push strategy when new products are put on the shelves, especially when there are many new products, how to accurately push information to improve operational effectiveness and reduce costs is an urgent problem.
By obtaining historical data and characteristic data of target product information, using time series prediction algorithm and enhancement tree model to calculate the popularity prediction score, predict success probability value and alternative score, comprehensively calculate the target comprehensive score, and determine the push of target product information Strategy.
The potential estimate of each target product information is achieved, accurate push, improved operational effectiveness, reduced operational costs, and improved customer experience.
Smart Images

Figure CN118941348A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of data processing and information push, and in particular to an information push method, device, electronic device and medium. Background Art
[0002] With the development of the Internet, e-commerce platforms have become mainstream media due to their advantages such as intelligence and convenience, and have a large amount of customer resources and information. In related technologies, when e-commerce platforms have new products on the shelves, they need to promote new product information to customers, but there are often many new products on the shelves in terms of quantity and variety. How to determine the operation and push strategy for each new product is an urgent problem to be solved. Summary of the invention
[0003] The present application aims to solve one of the technical problems in the related art at least to some extent.
[0004] To this end, one purpose of the present application is to propose a method for pushing information, by obtaining target product information to be pushed; obtaining target historical product information and historical product information data of the target historical product information under the product information category to which the target product information belongs, and obtaining a popularity prediction score of the target product information based on the historical product information data; obtaining a prediction success probability value of the target product information based on feature data of the target product information, and obtaining an alternative score of the target product information based on the interaction data of the target product information; determining a target comprehensive score of the target product information based on the popularity prediction score, the prediction success probability value and the alternative score, and pushing the target product information based on the target comprehensive score.
[0005] The second objective of the present application is to provide an information pushing device.
[0006] The third objective of the present application is to provide an electronic device.
[0007] A fourth object of the present application is to provide a non-transitory computer-readable storage medium.
[0008] A fifth object of the present application is to provide a computer program product.
[0009] To achieve the above-mentioned purpose, the first aspect embodiment of the present application proposes a method for pushing information, which obtains the target product information to be pushed; obtains the target historical product information and the historical product information data of the target historical product information under the product information category to which the target product information belongs, and obtains the popularity prediction score of the target product information based on the historical product information data; obtains the prediction success probability value of the target product information based on the characteristic data of the target product information, and obtains the alternative option score of the target product information based on the interaction data of the target product information; determines the target comprehensive score of the target product information based on the popularity prediction score, the prediction success probability value and the alternative option score, and pushes the target product information based on the target comprehensive score.
[0010] According to one embodiment of the present application, obtaining target product information to be pushed includes: obtaining a candidate feature vector corresponding to the candidate product information, wherein the candidate feature vector includes at least an image vector and a text vector of the candidate product information; calculating the similarity between the candidate feature vector and each historical feature vector in a vector library to obtain the similarity between the candidate feature vector and each historical feature vector; in response to the similarity being less than a preset similarity threshold, determining the candidate product information as the target product information.
[0011] According to one embodiment of the present application, based on the historical product information data, multiple historical timestamps of the target historical product information and the historical popularity values corresponding to the historical timestamps are obtained; the length of the time series corresponding to the future set time period that needs to be predicted is obtained; the multiple historical timestamps, the historical popularity values and the length of the time series corresponding to the future set time period are input into a time series prediction algorithm to obtain the popularity prediction score of the target product information.
[0012] According to one embodiment of the present application, based on the feature data of the target product information, obtaining the predicted success probability value of the target product information includes: obtaining the feature data of the target product information, wherein the feature data includes the basic merchant information of the merchant to which the target product information belongs, the sales data corresponding to the target product information, the word-of-mouth data corresponding to the target product information, and the traffic data corresponding to the target product information; inputting the feature data into a pre-trained boosting tree model, and obtaining the predicted success probability value of the target product information output by the boosting tree model.
[0013] According to one embodiment of the present application, a training method for a boosted tree model includes: obtaining positive sample feature data of sampled positive sample information, and obtaining negative sample feature data of sampled negative sample information; training an initial boosted tree model according to the positive sample feature data and the negative sample feature data to obtain a trained boosted tree model.
[0014] According to one embodiment of the present application, the interactive data includes accessed data, promotion cost data and interactive task achievement data corresponding to the target product information, and the alternative item scores of the target product information are obtained according to the interactive data of the target product information, including: obtaining the accessed data of the target product information, and determining the potential item score of the target product information according to the accessed data; and / or determining the promotion cost item score of the target product information according to the promotion cost data of the target product information; and / or determining the task item score of the target product information according to the interactive task achievement data of the target product information; using the potential item score, the promotion cost item score and the task item score as the alternative item scores.
[0015] According to one embodiment of the present application, a target comprehensive score for target product information is determined based on the popularity prediction score, the prediction success probability value and the alternative item score, including: obtaining the weights corresponding to the popularity prediction score, the prediction success probability value and the alternative item score; performing weighted calculation on the popularity prediction score, the prediction success probability value and the alternative item score according to their respective corresponding weights to obtain the target comprehensive score.
[0016] According to an embodiment of the present application, pushing target product information according to the target comprehensive score includes: determining a target push plan for the target product information according to the target comprehensive score; and pushing the target product information according to the target push plan.
[0017] According to an embodiment of the present application, after obtaining the target product information to be pushed, it also includes: binding the target product information with a preset logo, and displaying the preset logo on a display page of the target product information.
[0018] To achieve the above-mentioned purpose, the second aspect embodiment of the present application proposes an information pushing device, including: a first acquisition module, used to acquire target product information to be pushed; a second acquisition module, used to acquire target historical product information and historical product information data of the target historical product information under the product information category to which the target product information belongs, and acquire the popularity prediction score of the target product information based on the historical product information data; a third acquisition module, used to acquire the predicted success probability value of the target product information based on the characteristic data of the target product information, and acquire the alternative option score of the target product information based on the interactive data of the target product information; an information pushing module, used to determine the target comprehensive score of the target product information based on the popularity prediction score, the predicted success probability value and the alternative option score, and push the target product information based on the target comprehensive score.
[0019] According to one embodiment of the present application, the first acquisition module is also used to: obtain a candidate feature vector corresponding to the candidate product information, wherein the candidate feature vector includes at least a picture vector and a text vector of the candidate product information; calculate the similarity between the candidate feature vector and each historical feature vector in the vector library to obtain the similarity between the candidate feature vector and each historical feature vector; in response to the similarity being less than a preset similarity threshold, determine the candidate product information as the target product information.
[0020] According to one embodiment of the present application, the second acquisition module is also used to: obtain multiple historical timestamps of the target historical product information and the historical heat values corresponding to the historical timestamps based on the historical product information data; obtain the length of the time series corresponding to the future set time period that needs to be predicted; input the multiple historical timestamps, the historical heat values and the length of the time series corresponding to the future set time period into the time series prediction algorithm to obtain the heat prediction score of the target product information.
[0021] According to one embodiment of the present application, the third acquisition module is also used to: obtain feature data of the target product information, wherein the feature data includes basic merchant information of the merchant to which the target product information belongs, sales data corresponding to the target product information, word-of-mouth data corresponding to the target product information, and traffic data corresponding to the target product information; input the feature data into a pre-trained boosting tree model, and obtain the predicted success probability value of the target product information output by the boosting tree model.
[0022] According to one embodiment of the present application, the third acquisition module is also used to: obtain positive sample feature data of sampled positive sample information, and obtain negative sample feature data of sampled negative sample information; train the initial boosting tree model according to the positive sample feature data and the negative sample feature data, and obtain the trained boosting tree model.
[0023] According to one embodiment of the present application, the interactive data includes accessed data, promotion cost data and interactive task achievement data corresponding to the target product information. The third acquisition module is also used to: obtain the accessed data of the target product information, and determine the potential item score of the target product information based on the accessed data; and / or determine the promotion cost item score of the target product information based on the promotion cost data of the target product information; and / or determine the task item score of the target product information based on the interactive task achievement data of the target product information; and use the potential item score, promotion cost item score and task item score as alternative item scores.
[0024] According to one embodiment of the present application, the information push module is also used to: obtain the weights corresponding to the heat prediction score, the prediction success probability value, and the alternative option score; and perform weighted calculation on the heat prediction score, the prediction success probability value, and the alternative option score according to their corresponding weights to obtain the target comprehensive score.
[0025] According to an embodiment of the present application, the information push module is further used to: determine a target push plan for target product information according to the target comprehensive score; and push the target product information according to the target push plan.
[0026] According to an embodiment of the present application, the information push device also includes: a display module, which is used to bind the target product information with a preset logo and display the preset logo on a display page of the target product information.
[0027] To achieve the above-mentioned purpose, the third aspect embodiment of the present application proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to implement the information push method as described in the first aspect embodiment of the present application.
[0028] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the information push method as described in the first aspect embodiment of the present application.
[0029] To achieve the above-mentioned purpose, the fifth aspect embodiment of the present application proposes a computer program product, including a computer program, which, when executed by a processor, implements the information pushing method as described in the first aspect embodiment of the present application.
[0030] This application achieves at least the following beneficial effects:
[0031] This application realizes the potential estimation of each target product information to determine the operation resource supply strategy, realize the accurate push of target product information, improve the operation effect, and improve the customer experience while reducing the operation cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0033] Figure 1 It is a schematic diagram of an exemplary implementation of a method for pushing information shown in an embodiment of the present application.
[0034] Figure 2It is a schematic diagram of an exemplary implementation of a method for pushing information shown in an embodiment of the present application.
[0035] Figure 3 It is an overall framework diagram of a method for pushing information shown in an embodiment of the present application.
[0036] Figure 4 A heat trend curve chart based on a time series is shown in an embodiment of the present application.
[0037] Figure 5 It is a schematic diagram of a change point shown in an embodiment of the present application.
[0038] Figure 6 It is a schematic diagram showing an embodiment of the present application for obtaining a target comprehensive score and distinguishing product levels based on the target comprehensive score.
[0039] Figure 7 It is a schematic diagram of a product information submission and identification process shown in an embodiment of the present application.
[0040] Figure 8 It is a schematic diagram of an information pushing device shown in an embodiment of the present application.
[0041] Fig. 9 It is a schematic diagram of an electronic device shown in one embodiment of the present application. DETAILED DESCRIPTION
[0042] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0043] Figure 1 is a schematic diagram of an exemplary implementation of a method for pushing information shown in the present application, such as Figure 1 As shown, the method for pushing the information includes the following steps:
[0044] S101, obtaining target product information to be pushed.
[0045] Get the target product information to be pushed.
[0046] Optionally, the target product information may be new product information submitted by the merchant to the platform.
[0047] Optional. Since the new product information submitted by merchants to the platform is not necessarily a new product, other merchants may have already submitted this product on the platform. When the platform supports new products, in order to avoid wasting resources, it is necessary to screen the new product information submitted by merchants and use the new product information that meets the new product conditions as the target product information.
[0048] S102, obtaining target historical product information and historical product information data of the target historical product information under the product information category to which the target product information belongs, and obtaining a popularity prediction score of the target product information according to the historical product information data.
[0049] The product information category to which the target product information belongs, as well as the target historical product information under the product information category and the historical product information data of the target historical product information are obtained, wherein the historical product information data includes the sales data, click-through rate data, favorable comment data, etc. of the target historical product information. For example, if the target product information is a down jacket, the product information category to which it belongs is clothing, then various historical clothing information under the clothing category and the historical data corresponding to the historical clothing information are obtained.
[0050] Based on the historical product information data, multiple historical timestamps of the target historical product information and the historical popularity values corresponding to the historical timestamps are obtained, and the length of the time series corresponding to the future set time period that needs to be predicted is obtained. The multiple historical timestamps, historical popularity values and the length of the time series corresponding to the future set time period are input into the time series prediction algorithm to obtain the output popularity prediction score of the target product information within the future set time period.
[0051] S103, obtaining a prediction success probability value of the target product information according to the characteristic data of the target product information, and obtaining an alternative score of the target product information according to the interaction data of the target product information.
[0052] According to the characteristic data of the target product information, the prediction success probability value of the target product information is obtained. The prediction success probability value of the target product information can be understood as the probability value that the ranking rate of the target product's future sales in the category to which it belongs reaches n%, wherein the value of n can be set according to the actual situation, for example, n can be 5. Optionally, the characteristic data includes the merchant basic information of the merchant to which the target product information belongs, the sales data corresponding to the target product information, the word-of-mouth data corresponding to the target product information, and the traffic data corresponding to the target product information.
[0053] The alternative item scores of the target product information are obtained according to the interactive data of the target product information. Among them, the alternative items can be set according to the actual situation. For example, the interactive data may include the accessed data corresponding to the target product information (such as the number of independent visitors, click-through rate and other parameters), the promotion cost data (the amount of advertising money invested in the same category) and the interactive task achievement data (main image video, picture and text promotion and other tasks). Then, the potential item score of the target product information can be determined according to the accessed data of the target product information and the accessed data; the promotion cost item score of the target product information can be determined according to the promotion cost data of the target product information; the task item score of the target product information can be determined according to the interactive task achievement data of the target product information, and the potential item score, promotion cost item score and task item score are used as the alternative item score.
[0054] S104, determining a target comprehensive score of the target product information according to the popularity prediction score, the prediction success probability value and the alternative option score, and pushing the target product information according to the target comprehensive score.
[0055] According to the popularity prediction score, prediction success probability value and alternative score, the target comprehensive score of the target product information is determined. Optionally, the sum of the popularity prediction score, prediction success probability value and alternative score can be used as the target comprehensive score of the target product information. Optionally, corresponding weights can be set for the popularity prediction score item, prediction success probability value item and alternative score item respectively, and the weighted average is performed according to the popularity prediction score, prediction success probability value and alternative score and their corresponding weights to obtain the target comprehensive score of the target product information.
[0056] After obtaining the target comprehensive score of the target product information, compare the target comprehensive score with the preset candidate score range to obtain the target score range to which the target comprehensive score belongs, use the target push plan corresponding to the target score range as the target push plan for the target information, and push the target product information according to the target push plan.
[0057] The embodiment of the present application proposes a method for pushing information, by obtaining the target product information to be pushed; obtaining the target historical product information and the historical product information data of the target historical product information under the product information category to which the target product information belongs, and obtaining the heat prediction score of the target product information according to the historical product information data; obtaining the prediction success probability value of the target product information according to the characteristic data of the target product information, and obtaining the alternative score of the target product information according to the interaction data of the target product information; determining the target comprehensive score of the target product information according to the heat prediction score, the prediction success probability value and the alternative score, and pushing the target product information according to the target comprehensive score. The present application realizes the potential estimation of each target product information, so as to determine the operation resource supply strategy, realize the accurate push of the target product information, improve the operation effect, and improve the customer experience while reducing the operation cost.
[0058] Figure 2 is a schematic diagram of an exemplary implementation of a method for pushing information shown in the present application, such as Figure 2 As shown, the method for pushing the information includes the following steps:
[0059] S201, obtaining a candidate feature vector corresponding to the candidate product information, wherein the candidate feature vector at least includes a picture vector and a text vector of the candidate product information.
[0060] Figure 3 This is an overall framework diagram corresponding to a method for pushing information shown in this application, such as Figure 3 As shown in the figure, the framework as a whole includes a core algorithm module, a new product submission system module, a merchant-side kanban operation system module, and a resource operation module, which realizes the capabilities of new product image and text verification and recognition after new product submission, seasonal trend mining, potential new product prediction and stratification, new product operation strategy, merchant operation guidance, etc.
[0061] Among them, the merchant-side dashboard operation system module can include a current situation insight unit, a support strategy unit and an operation task unit, which are used to provide merchants with intuitive operation guidance and current situation insights, guide merchants to operate effectively according to current support strategies, promote merchant operation investment, achieve an ecological cycle, and help new products grow.
[0062] The image and text feature extraction model mainly converts the main image, title and core attribute information of the candidate product information into vector representation as the candidate feature vector corresponding to the candidate product information. Each candidate product information corresponds to a candidate feature vector, and the candidate feature vector includes at least the image vector and text vector of the candidate product information.
[0063] Among them, the image-text feature extraction model adopts the pre-trained Chinese cross-modal pre-training model (Contrastive Language–Image Pre-training, CLIP) architecture based on contrastive learning, and uses text as a supervisory signal to train a transferable visual model. The whole model includes two models: Text Encoder and Image Encoder. The Text Encoder is used to extract text features, using the optimized Bidirectional Encoder Representation from Transformers (BERT) method; while the Image Encoder is used to extract image features, using the ResNet of the Convolutional Neural Network (CNN) architecture.
[0064] Among them, in the training stage of the feature extraction algorithm model, the model training data is the main image, title and core attribute words of the platform's old products, using image (product main image)-text (corresponding title plus attribute) description, positive samples are a pair of image and text data belonging to the same product, and negative samples are not the same product. The training goal is to maximize the similarity of positive samples and minimize the similarity of negative samples. After large-scale data training, Chinese encoders and image encoders are generated, and the deep learning development framework Pytorch is used to build a feature extraction service module to achieve feature extraction and vectorization of product images and texts.
[0065] S202, calculating the similarity between the candidate feature vector and each historical feature vector in the vector library to obtain the similarity between the candidate feature vector and each historical feature vector.
[0066] The vector retrieval service mainly calculates the similarity between the candidate feature vector and each historical feature vector in the vector library corresponding to the old product, and obtains the similarity between the candidate feature vector and each historical feature vector.
[0067] After the similarity calculation between the candidate feature vector and each historical feature vector in the vector library is completed, the candidate feature vector can be added to the vector library.
[0068] S203: In response to the similarity being less than a preset similarity threshold, determining the candidate product information as target product information.
[0069] If the similarity is less than a preset similarity threshold, the candidate product information is determined as the target product information.
[0070] S204, obtaining target historical product information and historical product information data of the target historical product information under the product information category to which the target product information belongs, and obtaining a popularity prediction score of the target product information according to the historical product information data.
[0071] Figure 3 The trend item heat prediction model in the is used to obtain the product information category to which the target product information belongs, as well as the target historical product information under the product information category and the historical product information data of the target historical product information. According to the historical product information data, multiple historical timestamps of the target historical product information and the historical heat values corresponding to the historical timestamps are obtained, and the length of the time series corresponding to the future set time period to be predicted is obtained. The multiple historical timestamps, historical heat values and the length of the time series corresponding to the future set time period are input into the time series prediction algorithm to obtain the output heat prediction score of the target product information.
[0072] Among them, the time series prediction algorithm is based on an improved prophet algorithm model that includes time series decomposition and machine learning fitting technology. In addition to the popularity trend parameter term, season parameter term, holiday parameter term and error parameter term, the time series decomposition here takes into account the effects of holidays and e-commerce marketing rhythm, and improves the change point setting method in the model trend term.
[0073] Time series formula: y(t)=(t)+(t)+h()+∈ t .
[0074] In the above formula, y(t) represents the popularity prediction score of the target product information, g(t) represents the popularity trend parameter item, s(t) represents the season parameter item, h(t) represents the holiday parameter item, ∈ t Represents the error parameter term.
[0075] Figure 4 This application shows a heat trend graph based on a time series, such as Figure 4 As shown in the figure, the heat trend parameter g(t) reflects the changing trend of the time series in the non-periodic aspect. The segmented logistic regression growth model is used. In the time series of the training data, the trend of the curve changes at a specific time or with a potential cycle, so a change point setting is required. The change point detection method does not use the default parameters of the model, but is improved to an optimization problem. The variance is used as the loss function. Under the condition of satisfying the constraints, the BinSeg sequential greedy algorithm is used to search all the cutting positions, and the cutting method with the smallest loss function value is selected. Figure 5 is a schematic diagram of a change point shown in this application, such as Figure 5 As shown, t1 * , t2 * These are the two change points in the time series.
[0076] Among them, the seasonal parameter term s(t) captures seasonal changes based on the Fourier transform of sine and cosine functions.
[0077] Among them, the holiday parameter item h(t) represents the holiday item. In addition to daily holidays, dates that are in line with the rhythm of e-commerce marketing can also be set.
[0078] Among them, the error parameter term ∈ t Indicates other parameters that affect the popularity prediction score of the target product information.
[0079] S205: Obtain a prediction success probability value of the target product information according to the characteristic data of the target product information.
[0080] Figure 3 The potential prediction model in the paper draws on the stacking idea. By building a two-layer model of "boosting tree model + regression model", the popularity prediction score is integrated into the regression model to predict the success probability value of new product prediction and divide it into levels.
[0081] Among them, the boosted tree model adopts the xgboost classification model, takes gradient boost as the framework, selects classification and regression tree (CART) as the integrated machine learning algorithm of the base learner, uses the forward step-by-step algorithm to optimize the objective function, and determines the branch features and split points through the Approximate Algorithm, thereby completing the construction of the model. Exemplarily, the training data of the boosted tree model uses the ranking rate of new products in the category to which the future sales volume reaches 5% (i.e., successful new products) as positive samples, and the rest as negative samples. Obtain the positive sample feature data of the sampled positive sample information, and obtain the negative sample feature data of the sampled negative sample information; train the initial boosted tree model according to the positive sample feature data and the negative sample feature data to obtain the trained boosted tree model.
[0082] Among them, for the positive sample feature data and the negative sample feature data, the feature data mainly include the basic merchant information of the merchant to which the sample belongs (merchant level, merchant new product status, merchant main categories, brand scale and capabilities, store historical new product investment, product information quality, logistics services, after-sales services, dispute response and other parameters), the sales data corresponding to the sample (orders, sales volume, transaction amount, fan conversion, member conversion, store repurchase, purchasing users and other parameters), the word-of-mouth data corresponding to the sample (evaluation, good reviews and other parameters) and the traffic data corresponding to the sample (number of independent visitors, page views, searches and other parameters).
[0083] Acquire feature data of the target product information, wherein the feature data includes basic merchant information of the merchant to which the target product information belongs, sales data corresponding to the target product information, word-of-mouth data corresponding to the target product information, and traffic data corresponding to the target product information, input the feature data into a pre-trained boosting tree model, and obtain the predicted success probability value of the target product information output by the boosting tree model.
[0084] S206: Obtain the candidate score of the target product information according to the interactive data of the target product information.
[0085] The candidate scores of the target product information are obtained according to the interaction data of the target product information.
[0086] Exemplarily, the interaction data includes access data, promotion cost data, and interaction task achievement data corresponding to the target product information.
[0087] Then, the candidate scores of the target product information are obtained according to the interactive data of the target product information, including:
[0088] Obtain the access data of the target product information (such as the number of unique visitors, click-through rate and other parameters), and determine the potential item score of the target product information based on the access data.
[0089] Determine the promotion cost item score of the target product information based on the promotion cost data of the target product information (the amount of advertising money invested when belonging to the same category).
[0090] Determine the task item scores of the target product information based on the interactive task achievement data of the target product information (main image video, graphic and text promotion and other tasks).
[0091] The potential item score, promotion cost item score and task item score are used as alternative item scores.
[0092] It is not difficult to understand that the above-mentioned alternative scores can also be selected from one or more of the potential item scores, promotion cost item scores and task item scores, and it is not necessary for the potential item scores, promotion cost item scores and task item scores to exist at the same time.
[0093] S207, obtaining the weights corresponding to the popularity prediction score, the prediction success probability value, and the alternative option score.
[0094] Get the weights corresponding to the popularity prediction score, prediction success probability value, and alternative score. The weights can be set according to the actual situation, and the rules can be freely configured and modified according to the business.
[0095] S208, performing weighted calculation on the popularity prediction score, prediction success probability value, and alternative option score according to their respective corresponding weights to obtain a target comprehensive score.
[0096] According to their corresponding weights, the popularity prediction score, prediction success probability value, and alternative option score are weightedly calculated to obtain the target comprehensive score.
[0097] Figure 6 This is a schematic diagram of obtaining a target comprehensive score and distinguishing product levels according to the target comprehensive score shown in the present application, such as Figure 6 As shown, based on the regression model, weighted calculation is performed according to the popularity prediction score, prediction success probability value, potential item score, promotion cost item score and task item score to obtain the target comprehensive score, and the level of the target product information is determined according to the target comprehensive score. The level can be set according to the actual situation. Figure 6 There are 5 levels in total, namely A level, B level, C level, D level and E level.
[0098] The regression model belongs to Figure 3 As part of the potential prediction model in the regression model, the regression model uses ridge regression to estimate the relationship between variables and the degree of influence of independent variables on dependent variables. In the case of multicollinearity, additional deviation is added when estimating, which effectively reduces variance and improves fit. The independent variables are selected as heat prediction score, prediction success probability value and alternative option score.
[0099] When training the regression model, the future sales and traffic acquisition conditions of the sample's product category are used as supervision values to train the weights of each variable factor in the regression model.
[0100] S209, determining a target push plan for target product information according to the target comprehensive score.
[0101] The level of the target product information is determined according to the target comprehensive score, and the target push plan of the target product information is determined according to the level of the target product information.
[0102] S210, pushing the target product information according to the target push plan.
[0103] Push the target product information according to the target push plan.
[0104] For example, if the target product information belongs to the A and B levels, the target product information can be displayed as the top new product mainly on the search page. If the target product information belongs to the C and D levels, the target product information can be displayed as the middle new product mainly on the recommendation page. The A level has a greater tilt of resources than other levels, which can be understood as more search exposure, and the same applies to other levels. If the target product information belongs to the E level, no strategic promotion is performed.
[0105] This application realizes the potential estimation of each target product information to determine the operation resource supply strategy, realize the accurate push of target product information, improve the operation effect, and improve the customer experience while reducing the operation cost.
[0106] Furthermore, after obtaining the target product information to be pushed, the method further includes: binding the target product information with a preset logo, and displaying the preset logo on a display page of the target product information. Figure 7 This is a schematic diagram of a product information submission and identification process shown in this application, such as Figure 7 As shown, after the merchant launches a new product, the new product label is declared in the new product submission system and enters the review stage. The review message is transmitted to the core algorithm service through the big data tool and platform. The candidate feature vector of the submitted candidate product information is obtained through the image and text feature extraction model, and then the similarity of the same category products in the old product vector library is retrieved through the vector retrieval service. If the similarity is greater than the set threshold, the candidate product information is judged to be a highly similar image and text product, and it is considered that it does not meet the new product marking conditions. If the similarity is less than the set threshold, it is judged that the candidate product information is not a highly similar image and text, and it is considered that it meets the new product label conditions. The candidate feature vector of the submitted candidate product information is stored in the old product vector library, and the corresponding review status is returned. The reason and labeling status are displayed in the submission system to realize automatic enrichment of the stock library and support subsequent new product submission verification.
[0107] Furthermore, the resource operation end adopts a hierarchical traffic support operation strategy at the C end (platform user end) according to the level of new product stratification and task completion, which is divided into three support levels: traffic support for core locations in each field, traffic support for key locations in each field, and natural traffic support for each field. At the P end (operation staff end), the resource operation end selects new products participating in marketing activities through new product stratification and new product indicators, and monitors the marketing performance of new products in subsequent links. Combined with the merchant's kanban operation system, merchants at the B end (merchant end) understand the status of new products, the level of new product stratification, the threshold gap between traffic support and marketing activities, etc., task to-do items, operate new products, expose products, and let merchants operate products in accordance with platform rules and have a clear purpose. The three ends of P, B, and C work together to achieve a virtuous ecological cycle for new product operations.
[0108] Figure 8 is a schematic diagram of an information push device shown in this application, such as Figure 8 As shown, the information pushing device 800 includes a first acquisition module 801, a second acquisition module 802, a third acquisition module 803 and an information pushing module 804, wherein:
[0109] The first acquisition module 801 is used to acquire target product information to be pushed;
[0110] The second acquisition module 802 is used to acquire the target historical product information and the historical product information data of the target historical product information under the product information category to which the target product information belongs, and acquire the popularity prediction score of the target product information according to the historical product information data;
[0111] The third acquisition module 803 is used to acquire the prediction success probability value of the target product information according to the characteristic data of the target product information, and acquire the candidate score of the target product information according to the interaction data of the target product information;
[0112] The information push module 804 is used to determine the target comprehensive score of the target product information according to the popularity prediction score, the prediction success probability value and the alternative option score, and push the target product information according to the target comprehensive score.
[0113] This device estimates the potential of each target product information to determine the operation resource supply strategy, achieve accurate push of target product information, improve operation results, and improve customer experience while reducing operation costs.
[0114] According to one embodiment of the present application, the first acquisition module 801 is also used to: obtain a candidate feature vector corresponding to the candidate product information, wherein the candidate feature vector includes at least a picture vector and a text vector of the candidate product information; calculate the similarity between the candidate feature vector and each historical feature vector in the vector library to obtain the similarity between the candidate feature vector and each historical feature vector; in response to the similarity being less than a preset similarity threshold, determine the candidate product information as the target product information.
[0115] According to one embodiment of the present application, the second acquisition module 802 is also used to: obtain multiple historical timestamps of the target historical product information and the historical heat values corresponding to the historical timestamps based on the historical product information data; obtain the length of the time series corresponding to the future set time period that needs to be predicted; input the multiple historical timestamps, the historical heat values and the length of the time series corresponding to the future set time period into the time series prediction algorithm to obtain the heat prediction score of the target product information. According to one embodiment of the present application, the third acquisition module 803 is also used to: obtain feature data of the target product information, wherein the feature data includes the basic merchant information of the merchant to which the target product information belongs, the sales data corresponding to the target product information, the word-of-mouth data corresponding to the target product information and the traffic data corresponding to the target product information; input the feature data into a pre-trained boosting tree model, and obtain the prediction success probability value of the target product information output by the boosting tree model.
[0116] According to one embodiment of the present application, the third acquisition module 803 is also used to: obtain positive sample feature data of sampled positive sample information, and obtain negative sample feature data of sampled negative sample information; train the initial boosting tree model according to the positive sample feature data and the negative sample feature data, and obtain the trained boosting tree model.
[0117] According to one embodiment of the present application, the interactive data includes accessed data, promotion cost data and interactive task achievement data corresponding to the target product information. The third acquisition module 803 is also used to: obtain the accessed data of the target product information, and determine the potential item score of the target product information based on the accessed data; and / or determine the promotion cost item score of the target product information based on the promotion cost data of the target product information; and / or determine the task item score of the target product information based on the interactive task achievement data of the target product information; and use the potential item score, promotion cost item score and task item score as alternative item scores.
[0118] According to one embodiment of the present application, the information push module 804 is also used to: obtain the weights corresponding to the heat prediction score, the prediction success probability value, and the alternative option score; and perform weighted calculations on the heat prediction score, the prediction success probability value, and the alternative option score according to their corresponding weights to obtain a target comprehensive score.
[0119] According to an embodiment of the present application, the information push module 804 is further used to: determine a target push plan for target product information according to the target comprehensive score; and push the target product information according to the target push plan.
[0120] According to an embodiment of the present application, the information push device 800 further includes: a display module 805, which is used to bind the target product information with a preset logo and display the preset logo on a display page of the target product information.
[0121] In order to implement the above embodiment, the present application embodiment also proposes an electronic device 900, such as Fig. 9 As shown, the electronic device 900 includes: a processor 901 and a memory 902 communicatively connected to the processor, the memory 902 stores instructions executable by at least one processor, and the instructions are executed by at least one processor 901 to implement the information push method shown in the above embodiment.
[0122] In order to implement the above embodiment, the embodiment of the present application also proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to implement the information push method shown in the above embodiment.
[0123] In order to implement the above embodiments, the embodiments of the present application further provide a computer program product, including a computer program, which implements the information push method shown in the above embodiments when executed by a processor.
[0124] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.
[0125] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0126] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0127] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A method for pushing information, characterized in that: include: Obtain target product information to be pushed; Obtaining target historical product information under the product information category to which the target product information belongs and historical product information data of the target historical product information, and obtaining a popularity prediction score of the target product information according to the historical product information data; According to the characteristic data of the target product information, a prediction success probability value of the target product information is obtained, and according to the interaction data of the target product information, an alternative score of the target product information is obtained; The target comprehensive score of the target product information is determined according to the popularity prediction score, the prediction success probability value and the alternative option score, and the target product information is pushed according to the target comprehensive score.
2. The method according to claim 1, characterized in that The step of obtaining target product information to be pushed includes: Acquire a candidate feature vector corresponding to the candidate product information, wherein the candidate feature vector includes at least a picture vector and a text vector of the candidate product information; Calculating the similarity between the candidate feature vector and each historical feature vector in the vector library to obtain the similarity between the candidate feature vector and each historical feature vector; In response to the similarity being less than a preset similarity threshold, the candidate product information is determined as the target product information.
3. The method according to claim 1 or 2, characterized in that: The obtaining the popularity prediction score of the target product information according to the historical product information data includes: According to the historical product information data, obtaining multiple historical timestamps of the target historical product information and historical heat values corresponding to the historical timestamps; Get the length of the time series corresponding to the future set time period that needs to be predicted; The multiple historical timestamps, the historical popularity values and the length of the time series corresponding to the future set time period are input into a time series prediction algorithm to obtain the popularity prediction score of the target product information.
4. The method according to claim 1 or 2, characterized in that: The obtaining, according to the characteristic data of the target product information, a predicted success probability value of the target product information comprises: Acquire the characteristic data of the target product information, wherein the characteristic data includes basic merchant information of the merchant to which the target product information belongs, sales data corresponding to the target product information, word-of-mouth data corresponding to the target product information, and traffic data corresponding to the target product information; The feature data is input into a pre-trained boosting tree model, and a prediction success probability value of the target product information output by the boosting tree model is obtained.
5. The method according to claim 4, characterized in that The training method of the boosted tree model comprises: Acquire positive sample feature data of sampled positive sample information, and acquire negative sample feature data of sampled negative sample information; The initial boosting tree model is trained according to the positive sample feature data and the negative sample feature data to obtain the trained boosting tree model.
6. The method according to claim 1 or 2, characterized in that: The interactive data includes access data, promotion cost data, and interactive task achievement data corresponding to the target product information. The step of obtaining the candidate score of the target product information according to the interactive data of the target product information includes: Acquire accessed data of the target product information, and determine a potential item score of the target product information based on the accessed data; and / or, Determine the promotion cost item score of the target product information according to the promotion cost data of the target product information; and / or, Determining the task item score of the target product information according to the interactive task achievement data of the target product information; The potential item score, the promotion cost item score and the task item score are used as the alternative item scores.
7. The method according to claim 1 or 2, characterized in that: The step of determining a target comprehensive score of the target product information according to the popularity prediction score, the prediction success probability value and the alternative item score includes: Obtain the weights corresponding to the popularity prediction score, the prediction success probability value, and the alternative option score; The popularity prediction score, the prediction success probability value, and the alternative option score are weightedly calculated according to the respective corresponding weights to obtain the target comprehensive score.
8. The method according to claim 1 or 2, characterized in that: The pushing of the target product information according to the target comprehensive score includes: Determine a target push plan for the target product information according to the target comprehensive score; The target product information is pushed according to the target push plan.
9. The method according to claim 1, characterized in that: After obtaining the target product information to be pushed, the method further includes: The target product information is bound to a preset logo, and the preset logo is displayed on a display page of the target product information.
10. An information push device, characterized in that: include: The first acquisition module is used to acquire the target product information to be pushed; A second acquisition module is used to acquire target historical product information under the product information category to which the target product information belongs and historical product information data of the target historical product information, and acquire a popularity prediction score of the target product information according to the historical product information data; A third acquisition module is used to acquire a prediction success probability value of the target product information according to the characteristic data of the target product information, and acquire an alternative score of the target product information according to the interaction data of the target product information; The information push module is used to determine the target comprehensive score of the target product information according to the popularity prediction score, the prediction success probability value and the alternative option score, and push the target product information according to the target comprehensive score.
11. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.
12. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-9.
13. A computer program product, comprising a computer program, which, when executed by a processor, implements the steps according to any one of claims 1 to 9.
