Sorting model constructing method, commodity object searching method, device and medium

Abstract

Provided in the embodiments of the present application are a sorting model constructing method, a commodity object searching method, a device and a medium. The method comprises: deleting search association data of a target language site from search association data so as to obtain target search association data; according to a set rule, selecting feature data from the target search association data to construct training data; constructing a sorting model; and, on the basis of the training data, training the sorting model to obtain a sorting model meeting a condition. The present application can eliminate the affects of target languages on data, thus making the target search association data of each language more balanced, and eliminating the affects of non-uniform data distribution of each site on small-traffic languages; and the present application obtains the sorting model supporting various languages, such that searches in various languages all can be accurately fed back.

Description

Sorting model modeling, commodity object search methods, equipment and media

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on June 16, 2023, with application number 202310723141.5 and application name “Sorting model modeling, commodity object search method, device and medium”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of computer technology, and in particular to a method for constructing a ranking model, a method for searching for commodity objects, an electronic device, and a storage medium. Background Art

[0003] Cross-border e-commerce websites provide cross-border e-commerce services, allowing users to purchase goods from different countries. Since cross-border e-commerce websites support the sale of goods between different countries, they also support different languages.

[0004] When users search on cross-border e-commerce websites, they typically filter products based on keywords they enter. However, since e-commerce websites support multiple languages, the amount of user and product data corresponding to different languages ​​varies significantly. This impacts the ranking model. Products corresponding to languages ​​with more data are generally ranked higher in search rankings and are more easily reported to users during searches. Products in lesser-known languages, however, tend to be ranked lower in search rankings due to their smaller data volumes, resulting in less feedback during searches. This, in turn, impacts product searches and sales for some users who sell products in lesser-known languages.

[0005] Summary of the Invention

[0006] The embodiments of the present application provide a method for constructing a ranking model to improve the accuracy of the ranking model.

[0007] Correspondingly, the embodiments of the present application also provide a method for commodity objects, an electronic device and a storage medium to ensure the implementation and application of the above system.

[0008] In order to solve the above problems, the present invention discloses a method for constructing a ranking model, which includes:

[0009] Deleting the search related data of the target language site from the search related data to obtain the target search related data;

[0010] Filtering feature data from the target search related data according to set rules to construct training data;

[0011] Build a ranking model;

[0012] The sorting model is trained based on the training data to obtain a sorting model that meets the conditions.

[0013] Optionally, before deleting the search-related data of the target language site from the search-related data and obtaining the target search-related data, the following steps may be further included:

[0014] Get search related data for each language site;

[0015] The data volume information corresponding to each language site is determined based on the search-related data, and the target language site is determined based on the data volume information.

[0016] Optionally, filtering feature data from the target search associated data according to set rules to construct training data includes:

[0017] Sampling information of target search related data based on each language site;

[0018] According to the adopted information, data of target features are collected from target search related data of each language site to construct training data.

[0019] Optionally, the sample information of the target search related data based on each language site includes:

[0020] Based on the target search related data of each language site, determine the data volume information and behavior information of each language site;

[0021] Based on the data volume information and the behavior information, sampling information of the target search-related data of each language site is determined.

[0022] Optionally, also include:

[0023] Extracting feature data of each feature from the associated data, and analyzing correlation information between each feature data and the behavior information;

[0024] Target features are filtered based on the correlation information.

[0025] Optionally, the method further includes: constructing a site difference feature as a target feature for each commodity object.

[0026] Optionally, constructing site difference features for each commodity object includes:

[0027] Generate a differential feature based on at least one of a language feature, a product object feature, and a conversion feature;

[0028] Determine the site characteristics of the site corresponding to the product object;

[0029] The difference feature and the site feature are combined to obtain the corresponding site difference feature.

[0030] Optionally, generating a difference feature based on at least one of a language feature, a product object feature, and a conversion feature includes:

[0031] Obtaining the associated features of the product object, including the interaction depth feature of the language, the conversion rate feature of the product object, and the service quality feature of the seller user;

[0032] determining a conversion feature of the commodity object based on the associated feature;

[0033] Combine conversion features, language features, and product object features to determine corresponding difference features.

[0034] Optionally, constructing a ranking model includes:

[0035] An activation function is determined, and the activation function is added to the neural network model to construct a sorting model.

[0036] Optionally, determining the target language site based on the data volume information includes:

[0037] When the data volume information corresponding to the site exceeds the target threshold, the site is determined to be a target language site, and the language of the site is the target language.

[0038] The present application also discloses a method for searching for commodity objects, the method comprising:

[0039] receiving a search request, wherein the search request carries a keyword;

[0040] Searching based on the keywords to determine multiple product objects;

[0041] Analyzing multiple product objects based on a ranking model to select product objects, wherein the ranking model is obtained based on training data, the training data is constructed based on feature data selected according to set rules in target search related data, and the target search related data is determined by deleting the target language site from the search related data;

[0042] Generate search results based on the filtered product objects and provide feedback.

[0043] Optionally, analyzing the plurality of commodity objects based on the ranking model and screening the commodity objects includes:

[0044] Collect target feature data from the feature data of the product object and input it into the ranking model;

[0045] The product objects are screened based on the output result of the ranking model.

[0046] An embodiment of the present application also discloses an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method described in the embodiment of the present application.

[0047] The embodiment of the present application further discloses a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the method described in the embodiment of the present application.

[0048] Compared with the prior art, the embodiments of the present application have the following advantages:

[0049] In an embodiment of the present application, the search-related data of the target language site is deleted from the search-related data to obtain the target search-related data, thereby removing the influence of the target language on the data and making the target search-related data of each language more balanced. Then, feature data is filtered from the target search-related data according to set rules to construct training data. The required feature data can be filtered as training data, and then a sorting model is constructed. The sorting model is trained based on the training data to obtain a sorting model that meets the conditions, thereby eliminating the uneven distribution of data on each site for low-traffic languages ​​and obtaining a sorting model that supports various languages, so that searches in various languages ​​can be accurately fed back. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] FIG1 is a flowchart of an embodiment of a method for constructing a ranking model of the present application;

[0051] FIG2 is a flowchart of another embodiment of a method for constructing a ranking model of the present application;

[0052] FIG3 is a flowchart of an embodiment of a method for searching for commodity objects of the present application;

[0053] FIG4 is a schematic structural diagram of an exemplary device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0055] The embodiments of the present application can be applied to the search field. Cross-border e-commerce websites provide services to sellers and buyers in hundreds of countries and multiple languages. Therefore, websites can be set up in different countries for easy access. Different countries use different languages. For example, English-speaking countries such as the United Kingdom and the United States use English, and these websites can be called English sites. German websites use German, and the corresponding websites can be called German sites. Therefore, cross-border e-commerce websites include sites in various languages.

[0056] When training a sorting model, the embodiments of the present application can reduce the impact of big data language on sorting and improve the accuracy of the sorting model.

[0057] Step 102: Delete the search-related data of the target language site from the search-related data to obtain the target search-related data.

[0058] Search-related data is obtained from the database of a cross-border e-commerce website. This search-related data includes various data related to the searched product objects, such as product object data, keywords corresponding to the product objects, search click data, store data, etc. Therefore, search-related data for sites in various languages ​​can be obtained from the database of a cross-border e-commerce website. The language of a site is primarily based on the default language of the site. For example, sites for the UK and US are in English, sites for Japan are in Japanese, and sites for South Korea are in Korean.

[0059] The embodiment of the present application constructs a ranking model that supports searches in various languages. Among them, some language sites have more data in cross-border e-commerce websites. If the ranking model is to be trained for all languages ​​together, this more data will affect the accuracy of the ranking model, resulting in inaccurate ranking results. Therefore, the language with more data can be determined as the target language, the search-related data of the target language site can be determined, and the search-related data of the target language site can be deleted from the search-related data. In the embodiment of the present application, the search-related data of the target language site is deleted, not all the data of the target language mentioned above. There will also be some search data corresponding to the target language in sites of other languages, which will not affect the search ranking of the target language. For example, on a Korean site in South Korea, there will also be users who use English to search, that is, the language of the site is the default language of the site, and there will be no restrictions on the language used by users on the site. The embodiment of the present application can set the corresponding language as the target language, and the site with the target language as the base language is the target language site, such as an English site. In other embodiments, one or more target language sites can also be determined based on the amount of data corresponding to the site.

[0060] In an optional embodiment, the search-related data of the target language site is deleted from the search-related data, and before the target search-related data is obtained, the method further includes: obtaining the search-related data of each language site; determining the data volume information corresponding to each language site based on the search-related data, and determining the target language site based on the data volume information. The search-related data of each language site is obtained based on log data, etc., and then the data volume information corresponding to each language site, such as the data volume size, data volume proportion, etc., can be determined based on the search-related data. The target language site is determined based on the data volume information. The target language is usually one language, or more than one language, and is determined based on the data of each language site in the cross-border e-commerce website. In the embodiment of the present application, determining the target language site based on the data volume information includes: judging whether the data volume information corresponding to the language site exceeds the target threshold, and if the data volume information corresponding to a certain language site exceeds the target threshold, determining the language site as the target language site. Among them, the target threshold is the threshold for judging the target site, which can be a threshold for the size of the data volume or a threshold for the proportion of the data. For example, if the data volume of a certain site accounts for more than 60% of the total site data volume, or exceeds 50%, 40%, etc., it can be determined as a target language site.

[0061] In an example cross-border e-commerce website, the historical logs of the English site account for nearly 65% ​​of the entire website; while the low-traffic sites such as Korean, Japanese, and Vietnamese account for only less than 2%. If the data of each language site is used for unified training, the characteristics of the language sites of other small languages ​​may be completely dominated by the English site. Therefore, in the embodiment of the present application, the search-related data of the English site is deleted, and the search-related data of other language sites are used for training. It should be noted that in the data traffic of each language site other than English, there is still nearly 30% of search data corresponding to English keywords, which can make up for the missing proportion of the search-related data of the English site, so that the ranking model supporting each language still has the ability to learn English keywords.

[0062] Step 104 : filtering feature data from the target search associated data according to set rules to construct training data.

[0063] The target search associated data includes various data related to the search. However, not all data has an impact on the ranking, and different types of data have different effects on the ranking. Therefore, it is necessary to screen the required feature data and construct training samples.

[0064] Because training data is obtained from various language sites, and different language sites correspond to different traffic, user behavior, and other factors, it is also necessary to determine sampling information, such as the sampling ratio, for each language site's corresponding data. Training data is then sampled according to this sampling information. Feature data is filtered from the target search-related data according to set rules to construct training data. This includes: sampling information based on the target search-related data for each language site; and collecting target feature data from the search data for each language site according to the adoption information to construct training data. The sampling information based on the target search-related data for each language site includes: determining data volume and behavior information for each language site based on the target search-related data for each language site; and determining sampling information for the target search-related data for each language site based on this data volume and behavior information.

[0065] The data corresponding to different language sites varies. Some language sites have high traffic, meaning many users visit, while others have low traffic, meaning less data. However, the amount of traffic isn't directly proportional to the actual user interaction behavior on the site. Some sites have low traffic but high conversion rates for product objects. The conversion rate refers to the ratio of the number of users who use a single interactive behavior and overall functionality within a product object to the number of users who use it before and after. Therefore, based on the target search-related data for each language site, data volume information, such as the data volume percentage, can be determined for each language site. Conversion rate information corresponding to each language site can also be determined based on user behavior information. Then, based on the data volume and conversion rate information, sampling information, such as the sampling ratio, can be determined for each language site. Therefore, embodiments of the present application can adjust the sampling ratio for deep inquiry samples for different sites. An inquiry refers to the process of verbally or in writing, in order to explore transaction conditions for the purpose of buying or selling a product object. On e-commerce websites, a buyer's visit to a product object, adding to a purchase, and communication with a customer can all be referred to as an inquiry.

[0066] Taking a cross-border e-commerce website as an example, based on the analysis of website log data, it can be found that the data distribution of each site is also different. Among the sites in other languages ​​​​except English sites, the top five sites in terms of traffic are Spanish ( ES) sites, French (French, FR) sites, Russian (Russian, RU) sites, Arabic (Arabic, AR) sites, and Turkish (Turkish, TR) sites contribute nearly 60% of the traffic, especially the Spanish ES site, which accounts for 30% of the total traffic. However, from the perspective of user conversion, the conversion rate of sites such as Korean sites, Japanese sites, and Vietnamese sites, which account for less than 5%, is higher than the conversion rate of the top five sites in terms of traffic. In order to make the contribution of sites in different languages ​​more even, different sampling ratios of deep inquiry samples can be adjusted for different sites. For example, sites with low conversion rates of head traffic are sampled at a lower ratio, while sites with high conversion rates of tail traffic are sampled at a higher ratio, which can help alleviate the problem of sparse and long-tail behavioral data.

[0067] After determining sampling information such as the sampling ratio, we can then collect data on target features from search data across various language sites based on this information to construct training data. Since different data have different impacts on ranking, we can also analyze and determine target features and collect data for these features.

[0068] In an optional embodiment, feature data for each feature is extracted from the associated data, and the correlation between each feature data and behavioral information is analyzed. Target features are then filtered based on this correlation information. Behavioral information can include information about interactions with product objects, such as clicks, visits, add-to-carts, and purchases. The correlation between each feature data and behavioral information can be analyzed. For example, the payment time and brand age of a product order are weakly correlated features. Removing these weakly correlated features can reduce the impact of irrelevant features. For example, a drop-rank method can be used to assign importance scores to various features of the interaction behavior, with scores ranging from 0 to 1. The importance scores are then arranged along a scale from 0 to 1. Based on the scoring and processing logic, features with relatively small impact are then removed. The drop-rank method can reduce the probability of low-importance features appearing in the neural network by adding perturbation variables. This method considers the combination of features and is effective in filtering the top N features. It also requires minimal computation (only a single model training is required). Therefore, nearly 50% of weakly correlated features such as payment time and brand age can be deleted through the above method.

[0069] In the embodiment of the present application, keywords are also determined as target features based on historical data. It can obtain historical data within a specified time period, such as search keywords, such as 1 month, 3 months, half a year and other specified time periods, and then a language model (LM) such as uni, bi-gram and other models can be used to train and analyze the historical data to calculate the difference, conversion rate, popularity and score between keywords, and obtain keywords corresponding to the core attributes of each industry and each site as target features. By adding keywords corresponding to these core attributes as target features, the proportion of important features in the training data can be increased, and the dominance of core features can be improved. By mining the core attributes of different language sites and industries based on user perspectives, the click-through rate of commodity objects in actual applications is +13%, which verifies the correctness of this idea.

[0070] In the embodiments of the present application, a site difference feature is constructed as a target feature for each product object. The site difference feature is composed of a site feature combined with a difference feature, for example, a site feature and a difference feature are concatenated to obtain a difference feature. The site feature is a feature of the site of the product object. The difference feature can be constructed and generated by the language feature, product object feature, and conversion feature corresponding to each product object. The language feature is a feature corresponding to the language, for example, a corresponding feature value or vector value is set for each language as the feature of the language. The product object feature is a feature of the product object and can be determined based on product object information, etc. The conversion feature is a feature derived from multiple features of the product object. It is a feature obtained by integrating multiple features, for example, based on multiple information such as language, conversion rate, and service quality to determine the corresponding conversion feature. In an optional embodiment, associated features of the product object can be obtained, including the interaction depth feature of the language, the conversion rate feature of the product object, and the service quality feature of the seller user. The conversion feature of the product object is determined based on the associated features; the conversion feature, language feature, and product object feature are concatenated to determine the corresponding difference feature. Among them, the interaction depth characteristics of a language can be determined based on the interaction data corresponding to the language of the language, such as user consultation data on product objects, etc., for example, based on the depth characteristics of communication in each field under each multilingual site, basic information on product communication under each multilingual site, etc.

[0071] By building language- and site-specific features, we can balance the learning of differentiated product object representations across each site, while also enabling unified learning of differentiated features across different sites using the same model. This optimizes both resource utilization and conversion efficiency.

[0072] Then, according to the sampling information, data of target features can be collected from the target search related data of each language site to construct training data.

[0073] Step 106: construct a ranking model.

[0074] To sort the search results for product objects, it is necessary to first construct a sorting model, wherein the sorting model can be constructed based on various sorting algorithms, neural network models, linear sorting models, tree models, deep learning models, etc. The sorting model can be determined based on click-through rate, visit rate, etc.

[0075] In an embodiment of the present application, the sorting model can be constructed based on a neural network model, so the activation function can be determined, and the activation function is added to the neural network model to construct the sorting model. In an embodiment of the present application, a variety of activation functions can be set, for example, the activation function includes a linear activation function and a nonlinear activation function: sigmoid, ReLU, ELU, Leaky ReLU, CReLU, relu6, SELU, softplus, etc. Among them, the activation function is a function added to the neural network to help the network learn complex patterns in the data. In neurons, the input is applied to the activation function after a series of weighted summations, which is similar to the neuron-based model in the human brain. The activation function ultimately determines whether to pass the signal and the content to be emitted to the next neuron. Therefore, the efficiency of the neural network can be improved by adjustment. Among them, sigmoid is a smoothed step function used for binary classification. ReLU (Rectified Linear Unit) is used to activate some neurons and increase sparsity. Leaky ReLU (ReLU with leakage), the difference between Leaky ReLU and ReLU is that the negative axis retains the constant leak, so when the input information is less than 0, the information is not completely lost. ELU is the Exponential Linear Unit. CReLU is the Concatenated Rectified Linear Units, and ReLU6 is a rectified linear unit with 6 as the value. It is mainly used to have good numerical resolution when the mobile terminal has low precision. SELU is the Scaled Exponential Linear Unit. The Softplus function can be regarded as a smoothed version of the Relu function, and its derivative is just the Logistic function. Although the Softplus function also has the characteristics of unilateral inhibition and wide excitation boundaries, it does not have sparse activation and will not sparse the model.

[0076] The sorting model includes an embedding layer, which can convert the input features into feature vectors. The embodiment of the present application can determine the embedding layer based on loading personalized vectors, correlation model learning, etc., so as to initialize the parameters of the model. For example, a personalized vector recall pre-trained identifier embedding layer can be loaded to perform vector conversion on the identifier. Load the text embedding layer trained by the language representation model (Bidirectional Encoder Representation from Transformers, BERT) to perform vector conversion on text fragments such as keywords. Load the embedding layer trained by the Graph Neural Network (GNN) to perform vector conversion on the corresponding features.

[0077] In one example, for the target features of the input embedding layer, text analysis can be used to determine the symbols and segment the text, identify each text segment, and then determine the tokens of the text segment and the identifiers of each token. The semantic information and position information of each token are analyzed and converted into corresponding feature vectors.

[0078] In the ranking model, each feature vector can be input into a deep neural network (DNN) layer for dimensionality reduction, and the resulting dimensionality-reduced vector can be input into an attention network. Based on experience with correlation optimization, embodiments of the present application can implement self-attention optimization of keywords (queries) and titles (titles) in the ranking model, resulting in stronger keyword-title representation and a certain degree of enhancement in semantic matching capabilities within the ranking model.

[0079] In an embodiment of the present application, an optimizer may be used during the training of the sorting model to improve the efficiency of the training, making the training more stable and the model more robust.

[0080] Step 108: Train the ranking model based on the training data to obtain a ranking model that meets the conditions.

[0081] The training data can be input into the ranking model to determine the ranking result. Then, a loss function is determined based on the ranking result and the training data. The ranking model parameters are adjusted based on the loss function and training is continued until a ranking model that meets the conditions is obtained.

[0082] In summary, the search-related data of the target language site is deleted from the search-related data to obtain the target search-related data, thereby removing the influence of the target language on the data and making the target search-related data of each language more balanced. Then, the feature data is filtered from the target search-related data according to the set rules to construct the training data. The required feature data can be filtered as the training data, and then a sorting model is constructed. The sorting model is trained based on the training data to obtain a sorting model that meets the conditions, thereby eliminating the uneven distribution of data on each site for small-traffic languages, and obtaining a sorting model that supports various languages, so that searches in various languages ​​can be accurately fed back.

[0083] Based on the above embodiments, the embodiments of the present application also provide a method for constructing a ranking model, which can delete the target language site data, delete weakly related features, add core key attributes, and eliminate the impact of uneven distribution of data from each site on low-traffic languages.

[0084] 2 , there is shown a flowchart of another embodiment of a method for constructing a ranking model of the present application.

[0085] Step 202: Obtain search related data for each language site.

[0086] Step 204: Determine the data volume information corresponding to each language site based on the search-related data.

[0087] Step 206: Determine whether the data volume information corresponding to the site exceeds the target threshold.

[0088] If yes, step 208 is executed; if no, the language of other sites is detected.

[0089] Step 208: Determine that the site is a target language site and the language of the site is the target language.

[0090] Step 210: Delete the search-related data of the target language site from the search-related data to obtain the target search-related data.

[0091] Step 212: Search for sample information of the associated data based on the target of each language site.

[0092] The data volume information and behavior information of each language site are determined based on the target search related data of each language site; and the sampling information of the target search related data of each language site is determined based on the data volume information and behavior information.

[0093] Step 214 : extracting feature data of each feature from the associated data, and analyzing the correlation information between each feature data and the behavior information.

[0094] Step 216: Filter target features based on the correlation information.

[0095] Step 218: For each commodity object, construct a site difference feature as a target feature.

[0096] The step of constructing a site difference feature for each product object includes: generating a difference feature based on at least one of a language feature, a product object feature, and a conversion feature; determining a site feature of a site corresponding to the product object; and combining the difference feature with the site feature to obtain a corresponding site difference feature.

[0097] The generating of difference features based on at least one of the language feature, the product object feature and the conversion feature includes: obtaining associated features of the product object, the associated features including the interaction depth feature of the language, the conversion rate feature of the product object and the service quality feature of the seller user; determining the conversion feature of the product object based on the associated features; and splicing the conversion feature, the language feature and the product object feature to determine the corresponding difference feature.

[0098] Step 220 : collecting target feature data from target search related data of each language site according to the adoption information to construct training data.

[0099] Step 222: determine an activation function, and add the activation function to the neural network model to construct a sorting model.

[0100] Step 224: Train the ranking model based on the training data to obtain a ranking model that meets the conditions.

[0101] In each embodiment of the present application, the order of constructing the ranking model and determining the training data is not limited, and they can also be determined in parallel and can be determined according to needs.

[0102] Compared with the problem of uneven training of small languages ​​in a single model in the unified modeling of the entire site, in the embodiment of the present application, the data of the target language site with a relatively large data proportion, such as the English site data, is removed, and the search data of the target language in the multilingual site is used to ensure that the model takes into account the target reasons. The municipal water conservancy in this application also adjusts the sampling ratio of different in-depth inquiry samples for different language sites, so that sites with a large amount of access data are set to sample data at a lower ratio, while sites with a low amount of access data are sampled at a higher ratio, which can help alleviate the problem of sparse and long-tail behavior data.

[0103] Compared with the problem of too many models caused by independent sorting modeling of sub-sites, which results in loss of resources and online performance. In addition, the amount of data in some low-traffic scenarios is small, and the model modeling is cross-modeled, which leads to serious loss of retention and conversion. The embodiment of the present application can delete weakly related features when determining training data, and add core key attributes. It can also construct differentiated features of language sites, eliminating the impact of uneven distribution of data at each site on the effect of low-traffic languages.

[0104] Compared with regional modeling, there is the problem of inconsistent division methods for different regions, and using translated English and Chinese information for matching will cause loss of semantic matching for different languages. The regions are close but the cultural customs are very different, resulting in the problem of lack of differentiation in site searches. Can the embodiment of the present application construct differentiated features of the site, so that the model can adapt to the language differences of each site when training. It also adjusts the activation function of the neural network in the model, improves the efficiency and learning ability of model training through the optimizer, and enhances the robustness of multilingual expression difference modeling.

[0105] After the ranking model is constructed, it can be used during website searches to improve ranking accuracy and enable searches in various languages ​​to obtain accurate and rich results.

[0106] 3 , there is shown a flowchart of steps of an embodiment of a method for searching for commodity objects according to an embodiment of the present application.

[0107] Step 302: Receive a search request, where the search request carries a keyword.

[0108] When a user browses a language site on a cross-border e-commerce website and wants to search for a product, they can enter a corresponding keyword. The keyword the user searches for on the website can be the primary language of the website or another language.

[0109] Step 304: Search based on the keyword to determine multiple product objects.

[0110] A search can be performed based on the search keyword, wherein the product objects searched can be product objects published in the language or product objects published in other languages. Multiple product objects can be searched, for example, more than 50 or more than 100.

[0111] Step 306: Analyze multiple product objects based on a ranking model and filter the product objects, wherein the ranking model is obtained based on training data, the training data is constructed based on feature data filtered according to set rules in the target search association data, and the target search association data is determined by deleting the target language site from the search association data.

[0112] The feature data of each product object can be determined, and target feature data can be collected from the feature data and input into a ranking model. Product objects can then be screened based on the output of the ranking model. For example, the ranking model can predict the click-through rate, visit rate, etc. of product objects based on the target feature and then rank them. The top K product objects are selected.

[0113] The feature data of each feature may be extracted from the associated data, and the correlation information between each feature data and the behavior information may be analyzed; and the target feature may be screened based on the correlation information.

[0114] For each product object, a site difference feature is constructed as a target feature. This construction of the site difference feature for each product object includes: generating a difference feature based on at least one of a language feature, a product object feature, and a conversion feature; determining a site feature for the site corresponding to the product object; and combining the difference feature with the site feature to obtain the corresponding site difference feature.

[0115] The generating of difference features based on at least one of the language feature, the product object feature and the conversion feature includes: obtaining associated features of the product object, the associated features including the interaction depth feature of the language, the conversion rate feature of the product object and the service quality feature of the seller user; determining the conversion feature of the product object based on the associated features; and splicing the conversion feature, the language feature and the product object feature to determine the corresponding difference feature.

[0116] Step 308: Generate search results based on the filtered product objects and provide feedback.

[0117] Generate search results based on the filtered K product objects and then feed back the search results.

[0118] In summary, a search request is received, the search request carries a keyword, and then a search is performed based on the keyword to determine multiple product objects. Then, the multiple product objects are analyzed based on a ranking model to filter the product objects, wherein the ranking model is obtained based on training data, and the training data is constructed based on feature data filtered according to set rules in target search related data. The target search related data is determined by deleting the target language site from the search related data, thereby supporting searches in various language sites, sorting product objects more accurately, generating search results based on the filtered product objects and providing feedback, providing users with accurate feedback results, and improving user experience.

[0119] It should be noted that the embodiments of the present application may involve the use of user data. In actual applications, user-specific personal data can be used in the scheme described herein within the scope permitted by applicable laws and regulations, subject to the requirements of applicable laws and regulations of the country where the user is located (for example, with the user's explicit consent, effective notification to the user, etc.).

[0120] It should be noted that for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the order of the actions described, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0121] Based on the above embodiment, this embodiment further provides a device for constructing a ranking model, which is applied to an electronic device such as a server device. The device includes:

[0122] A data filtering module is used to delete the search related data of the target language site from the search related data to obtain the target search related data;

[0123] A training data construction module is used to filter feature data from the target search related data according to set rules to construct training data;

[0124] Model building module, used to build ranking models;

[0125] The training module is used to train the sorting model based on the training data to obtain a sorting model that meets the conditions.

[0126] In summary, the search-related data of the target language site is deleted from the search-related data to obtain the target search-related data, thereby removing the influence of the target language on the data and making the target search-related data of each language more balanced. Then, the feature data is filtered from the target search-related data according to the set rules to construct the training data. The required feature data can be filtered as the training data, and then a sorting model is constructed. The sorting model is trained based on the training data to obtain a sorting model that meets the conditions, thereby eliminating the uneven distribution of data on each site for small-traffic languages, and obtaining a sorting model that supports various languages, so that searches in various languages ​​can be accurately fed back.

[0127] The data screening module is further configured to obtain search-related data of each language site; determine data volume information corresponding to each language site based on the search-related data; and determine a target language site based on the data volume information.

[0128] The training data construction module is used to collect target feature data from the target search associated data of each language site based on sampling information of the target search associated data of each language site according to the sampling information to construct training data.

[0129] The training data construction module is used to determine the data volume information and behavior information of each language site based on the target search related data of each language site; and determine the sampling information of the target search related data of each language site based on the data volume information and behavior information.

[0130] The training data construction module is further configured to extract feature data of each feature from the associated data, analyze correlation information between each feature data and the behavior information, and filter target features based on the correlation information.

[0131] The training data construction module is further used to construct site difference features as target features for each commodity object.

[0132] The training data construction module is further used to generate difference features based on at least one of language features, product object features and conversion features; determine the site features of the site corresponding to the product object; and splice the difference features and site features to obtain corresponding site difference features.

[0133] The training data construction module is also used to obtain associated features of product objects, which include language interaction depth features, product conversion rate features, and seller user service quality features; determine the conversion features of the product objects based on the associated features; and splice the conversion features, language features, and product object features to determine corresponding difference features.

[0134] The model building module is used to determine the activation function and add the activation function to the neural network model to build a sorting model.

[0135] The data screening module is configured to determine that the site is a target language site and the language of the site is the target language when the data volume information corresponding to the site exceeds a target threshold.

[0136] Compared with the problem of uneven training of small languages ​​in a single model in the unified modeling of the entire site, in the embodiment of the present application, the data of the target language site with a relatively large data proportion, such as the English site data, is removed, and the search data of the target language in the multilingual site is used to ensure that the model takes into account the target reasons. The municipal water conservancy in this application also adjusts the sampling ratio of different in-depth inquiry samples for different language sites, so that sites with a large amount of access data are set to sample data at a lower ratio, while sites with a low amount of access data are sampled at a higher ratio, which can help alleviate the problem of sparse and long-tail behavior data.

[0137] Compared with the problem of too many models caused by independent sorting modeling of sub-sites, which results in loss of resources and online performance. In addition, the amount of data in some low-traffic scenarios is small, and the model modeling is cross-modeled, which leads to serious loss of retention and conversion. The embodiment of the present application can delete weakly related features when determining training data, and add core key attributes. It can also construct differentiated features of language sites, eliminating the impact of uneven distribution of data at each site on the effect of low-traffic languages.

[0138] Compared with regional modeling, there is the problem of inconsistent division methods for different regions, and using translated English and Chinese information for matching will cause loss of semantic matching for different languages. The regions are close but the cultural customs are very different, resulting in the problem of lack of differentiation in site searches. Can the embodiment of the present application construct differentiated features of the site, so that the model can adapt to the language differences of each site when training. It also adjusts the activation function of the neural network in the model, improves the efficiency and learning ability of model training through the optimizer, and enhances the robustness of multilingual expression difference modeling.

[0139] Based on the above embodiment, this embodiment further provides a commodity object search device, which is applied to an electronic device such as a server device, and includes:

[0140] A receiving module, configured to receive a search request, wherein the search request carries a keyword;

[0141] A search module, configured to search based on the keywords and determine multiple commodity objects;

[0142] a ranking module, configured to analyze and select a plurality of product objects based on a ranking model, wherein the ranking model is obtained based on training data, the training data is constructed based on feature data selected according to set rules in target search-related data, and the target search-related data is determined by deleting a target language site from the search-related data;

[0143] The feedback module is used to generate search results and provide feedback based on the filtered product objects.

[0144] The sorting module is used to collect target feature data from the feature data of the commodity objects and input the data into the sorting model; and screen the commodity objects based on the output results of the sorting model.

[0145] In summary, a search request is received, the search request carries a keyword, and then a search is performed based on the keyword to determine multiple product objects. Then, the multiple product objects are analyzed based on a ranking model to filter the product objects, wherein the ranking model is obtained based on training data, and the training data is constructed based on feature data filtered according to set rules in target search related data. The target search related data is determined by deleting the target language site from the search related data, thereby supporting searches in various language sites, sorting product objects more accurately, generating search results based on the filtered product objects and providing feedback, providing users with accurate feedback results, and improving user experience.

[0146] An embodiment of the present application further provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of each method step in the embodiment of the present application.

[0147] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the embodiment of the present application.

[0148] The present application also provides an electronic device comprising: a processor and a memory communicatively connected to the processor; the memory storing computer-executable instructions; and the processor executing the computer-executable instructions stored in the memory to implement the method described in the present application. In the present application, the electronic device includes a server, a terminal device, and other devices.

[0149] The embodiments of the present disclosure may be implemented as a device configured as desired using any appropriate hardware, firmware, software, or any combination thereof, and the device may include electronic devices such as a server (cluster), a terminal, etc. FIG4 schematically illustrates an exemplary device 400 that may be used to implement various embodiments described in this application.

[0150] For one embodiment, Figure 4 shows an exemplary apparatus 400 having one or more processors 402, a control module (chip set) 404 coupled to at least one of the processor(s) 402, a memory 406 coupled to the control module 404, a non-volatile memory (NVM) / storage device 408 coupled to the control module 404, one or more input / output devices 410 coupled to the control module 404, and a network interface 412 coupled to the control module 404.

[0151] Processor 402 may include one or more single-core or multi-core processors, and processor 402 may include any combination of general-purpose processors or dedicated processors (e.g., graphics processors, application processors, baseband processors, etc.). In some embodiments, apparatus 400 can serve as a server, terminal, or other device described in the embodiments of the present application.

[0152] In some embodiments, the apparatus 400 may include one or more computer-readable media (e.g., memory 406 or NVM / storage 408) having instructions 414 and one or more processors 402 configured in conjunction with the one or more computer-readable media to execute the instructions 414 to implement a module to perform the actions described in the present disclosure.

[0153] For one embodiment, control module 404 may include any suitable interface controller to provide any suitable interface to at least one of processor(s) 402 and / or any suitable device or component in communication with control module 404 .

[0154] The control module 404 may include a memory controller module to provide an interface to the memory 406. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0155] The memory 406 can be used, for example, to load and store data and / or instructions 414 for the device 400. For one embodiment, the memory 406 can include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 406 can include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0156] For one embodiment, the control module 404 may include one or more input / output controllers to provide interfaces to the NVM / storage device 408 and the input / output device(s) 410 .

[0157] For example, NVM / storage 408 may be used to store data and / or instructions 414. NVM / storage 408 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0158] NVM / storage device 408 may include storage resources that are part of the device on which apparatus 400 is installed, or it may be accessible to the device without being part of the device. For example, NVM / storage device 408 may be accessible via a network via input / output device(s) 410.

[0159] (One or more) input / output devices 410 may provide an interface for apparatus 400 to communicate with any other appropriate devices. Input / output devices 410 may include communication components, audio components, sensor components, etc. Network interface 412 may provide an interface for apparatus 400 to communicate via one or more networks. Apparatus 400 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard such as Bluetooth, WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0160] For one embodiment, at least one of the processor(s) 402 may be packaged together with the logic of one or more controllers (e.g., a memory controller module) of the control module 404. For one embodiment, at least one of the processor(s) 402 may be packaged together with the logic of one or more controllers of the control module 404 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 402 may be integrated on the same die with the logic of one or more controllers of the control module 404. For one embodiment, at least one of the processor(s) 402 may be integrated on the same die with the logic of one or more controllers of the control module 404 to form a system-on-chip (SoC).

[0161] In various embodiments, apparatus 400 may be, but is not limited to, a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, apparatus 400 may have more or fewer components and / or a different architecture. For example, in some embodiments, apparatus 400 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0162] Among them, the main control chip can be used as a processor or control module in the detection device, sensor data, location information, etc. are stored in the memory or NVM / storage device, the sensor group can be used as an input / output device, and the communication interface may include a network interface.

[0163] The present application also provides an electronic device comprising: a processor; and a memory storing executable code, wherein when the executable code is executed, the processor executes one or more methods described in the embodiments of the present application. The memory in the embodiments of the present application can store various data, such as target files, file-application association data, and other data, and can also include user behavior data, thereby providing a data foundation for various processing.

[0164] The embodiments of the present application further provide one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the embodiments of the present application.

[0165] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0166] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0167] The present application embodiment is described with reference to the flow chart and / or block diagram of the method, terminal device (system), and computer program product according to the embodiment of the present application. It should be understood that each process and / or box in the flow chart and / or block diagram and the combination of the process and / or box in the flow chart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device produce a device for realizing the function specified in one process or multiple processes and / or one box or multiple boxes of the flow chart.

[0168] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0169] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce computer-implemented processing, so that the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0170] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0171] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0172] The above is a detailed introduction to a hand modeling method, a processing method based on a hand model, an electronic device and a storage medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for constructing a ranking model, characterized in that: The method comprises: Deleting the search related data of the target language site from the search related data to obtain the target search related data; From the target search related data, feature data is filtered according to set rules to construct training data; Build a sorting model; The sorting model is trained based on the training data to obtain a sorting model that meets the conditions.

2. The method according to claim 1, characterized in that Before deleting the search related data of the target language site from the search related data to obtain the target search related data, the following further includes: Get search related data of each language site; The data volume information corresponding to each language site is determined based on the search-related data, and the target language site is determined based on the data volume information.

3. The method according to claim 2, characterized in that The target search related data is searched for feature data according to set rules to construct training data, including: Sampling information of target search related data based on each language site; According to the adopted information, data of target features are collected from target search related data of each language site to construct training data.

4. The method according to claim 3, characterized in that The sample information of the target search related data based on each language site includes: Based on the target search related data of each language site, determine the data volume information and behavior information of each language site; Based on the data volume information and the behavior information, sampling information of the target search related data of each language site is determined.

5. The method according to claim 3, characterized in that: Also includes: Extracting feature data of each feature from the associated data, and analyzing the correlation information between each feature data and the behavior information; Target features are filtered based on the correlation information.

6. The method according to claim 3, characterized in that Also includes: For each commodity object, a site difference feature is constructed as the target feature.

7. The method according to claim 6, characterized in that The site difference features are constructed for each commodity object, including: Generate a difference feature based on at least one of a language feature, a commodity object feature, and a conversion feature; Determine the site characteristics of the site corresponding to the commodity object; The difference feature and the site feature are concatenated to obtain the corresponding site difference feature.

8. The method according to claim 7, characterized in that The generating of difference features based on at least one of the language feature, the commodity object feature and the conversion feature includes: Obtaining correlation features of the product object, the correlation features include the interaction depth feature of the language, the conversion rate feature of the product object, and the service quality feature of the seller user; Determining a conversion feature of the commodity object based on the association feature; Conversion features, language features, and product object features are combined to determine the corresponding difference features.

9. The method according to claim 1, characterized in that: The constructing of the sorting model comprises: An activation function is determined, and the activation function is added to the neural network model to construct a sorting model.

10. The method according to claim 2, characterized in that Determining a target language site based on the data volume information includes: When the data volume information corresponding to the site exceeds the target threshold, the site is determined to be a target language site, and the language of the site is the target language.

11. A commodity object search method, characterized in that: The method comprises: receiving a search request, wherein the search request carries a keyword; Searching based on the keyword to determine multiple commodity objects; Analyze multiple commodity objects based on a ranking model to filter commodity objects, wherein the ranking model is obtained based on training data, the training data is constructed based on feature data filtered according to set rules in target search related data, and the target search related data is determined by deleting a target language site from the search related data; Generate search results based on the filtered product objects and feedback.

12. The method according to claim 11, characterized in that The analyzing the multiple commodity objects based on the sorting model and screening the commodity objects includes: Collect target feature data from the feature data of the commodity object and input it into the sorting model; The commodity objects are screened based on the output result of the ranking model.

13. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 12.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 12 when executed by a processor.

Images