Method, device, electronic device and storage medium for sorting search attribute information

Abstract

An embodiment of the present application provides a method for sorting search attribute information, a device for sorting search attribute information, an electronic device and a storage medium. The method for sorting search attribute information includes: performing online recall processing based on a received search term of a product to obtain multiple recall attribute information corresponding to the search term; determining reference attribute information from candidate attribute information that meets association conditions with the search term screened out from an online file, wherein the reference attribute information is candidate attribute information selected for display, and the posterior probability of the reference attribute information is within a posterior probability threshold range; wherein the reference attribute information is partially or completely different from the multiple recall attribute information; and sorting the reference attribute information with the sorted multiple recall attribute information to obtain multiple ordered target attribute information that meets the sorting conditions and corresponds to the search term for display.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a method for sorting search attribute information, a device for sorting search attribute information, an electronic device, and a storage medium. Background Art

[0002] With the continuous improvement of Internet infrastructure and the popularization of computer network technology, the scale of merchants and commodities involved in e-commerce malls has gradually increased, and the attribute information of commodities has become more and more abundant. In this application scenario, after the user enters the search term (query) through the search engine in the mall, the search engine will filter and sort the attributes related to the search term. It is particularly important to ensure that the commodity information displayed based on the filtered attributes meets the user's search intention.

[0003] In the related art, the attribute screening module obtains an attribute list based on the search term through an offline mining strategy, and sorts the attributes in the list through the following calculation formula, and delays the update of the offline sorting result, i.e. the sorted attributes, to the search engine.

[0004]

[0005] Among them, n represents the total number of attribute features, w i represents the attribute feature weight of the i-th dimension, f i represents the attribute feature of the i-th dimension, and Normalized is a normalization processing function. However, this method has at least the problem that the offline sorting results obtained by the linear model are relatively stable, and there is a certain delay in updating the sorted attributes to the search engine, which leads to the problem that the Matthew effect caused by attribute screening is gradually increased. Summary of the invention

[0006] The embodiments of the present application are intended to provide a method for sorting search attribute information, a device for sorting search attribute information, an electronic device and a storage medium, so as to solve the problem in the related art that at least the offline sorting results obtained by the linear model are relatively stable, and there is a certain delay in updating the sorted attributes to the search engine, resulting in a gradual increase in the Matthew effect caused by attribute screening.

[0007] The technical solution of the embodiment of the present application is implemented as follows:

[0008] In a first aspect, an embodiment of the present application provides a method for sorting search attribute information, the method comprising:

[0009] Performing online recall processing based on the received search terms of the product to obtain multiple recall attribute information corresponding to the search terms;

[0010] Determining reference attribute information from candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information;

[0011] The reference attribute information and the sorted multiple recalled attribute information are sorted to obtain multiple ordered target attribute information that meet the sorting conditions and correspond to the search terms for display.

[0012] In a second aspect, an embodiment of the present application provides a device for sorting search attribute information, the device comprising:

[0013] A first processing module, configured to perform online recall processing based on the received search terms of the product, and obtain multiple recall attribute information corresponding to the search terms;

[0014] A second processing module is used to determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information;

[0015] A sorting module is used to sort the reference attribute information and the sorted multiple recalled attribute information to obtain multiple ordered target attribute information corresponding to the search term for display and meeting the sorting condition.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising: a processor, a memory, and a communication bus;

[0017] The communication bus is used to realize the communication connection between the processor and the memory;

[0018] The processor is used to execute the sorting program of the search attribute information stored in the memory to implement the sorting method of the search attribute information as described above.

[0019] In a fourth aspect, an embodiment of the present application provides a storage medium, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the above-mentioned method for sorting search attribute information.

[0020] The embodiments of the present application are applied to achieve the following beneficial effects: it is possible to add reference attribute information whose posterior probability is within the posterior probability threshold range to the multiple recalled attribute information after sorting, thereby avoiding the Matthew effect caused by attribute screening, and providing accurate and rich attribute sorting results for the attribute information in the attribute screening module, thereby maximizing the click-through rate and conversion rate of the screening, thereby shortening the user's browsing and purchasing path.

[0021] Because online recall processing is performed based on the received search terms of the product, multiple recalled attribute information corresponding to the search terms are obtained; reference attribute information is determined from the candidate attribute information that meets the association conditions with the search terms screened out from the online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information; the reference attribute information and the sorted multiple recalled attribute information are sorted to obtain multiple ordered target attribute information that meet the sorting conditions and correspond to the search terms for display; in this way, the problem that at least the offline sorting results obtained by the linear model are relatively stable and there is a certain delay in updating the sorted attributes to the search engine in the related art is solved, and the reference attribute information whose posterior probability is within the posterior probability threshold range is added to the sorted multiple recalled attribute information, thereby avoiding the Matthew effect caused by attribute screening, and providing accurate and rich attribute sorting results for the attribute information in the attribute screening module, maximizing the click-through rate and conversion rate of the screening, thereby shortening the user's browsing and purchasing path. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 An optional flowchart of a method for sorting search attribute information provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a recall logic for multiple recall attributes provided in an embodiment of the present application;

[0024] Figure 3 An optional flowchart of a method for sorting search attribute information provided in an embodiment of the present application;

[0025] Figure 4 An optional flowchart of a method for sorting search attribute information provided in an embodiment of the present application;

[0026] Figure 5 A schematic diagram of the structure of multiple ordered attribute information displayed by the reference attribute information and user feedback behavior data provided in the embodiment of the present application;

[0027] Figure 6An optional flowchart of a method for sorting search attribute information provided in an embodiment of the present application;

[0028] Figure 7 An optional flowchart of a method for sorting search attribute information provided in an embodiment of the present application;

[0029] Figure 8 An optional structural diagram of a device for sorting search attribute information provided in an embodiment of the present application;

[0030] Fig. 9 An optional structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0032] It should be understood that the "embodiments of the present application" or "the aforementioned embodiments" mentioned throughout the specification mean that specific features, structures or characteristics related to the embodiments are included in at least one embodiment of the present application. Therefore, "in the embodiments of the present application" or "in the aforementioned embodiments" appearing throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. In the various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned sequence numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0033] With the continuous improvement of Internet infrastructure and the continuous popularization of computer network technology, taking the search technology of online shopping as an example, in e-commerce search, users need to actively express their needs (users enter search terms) through the search box of the user interface, and the e-commerce platform also needs to provide guidance and screening functions, so that users can quickly find the products they want. As the scale of merchants and products in electronic malls gradually increases, and the attributes and label information of products are becoming more and more abundant, the attribute screening module in the screening function is particularly important. In an application scenario, the user enters the search term in the search box of the user interface, and the product search engine returns a product display list and the attributes included in the attribute screening module. At this time, the user accurately screens out the products he wants according to the attribute sorting results displayed in the attribute screening module to shorten the browsing and purchasing path. The specific implementation method of this application will be explained below using the search of products as an example.

[0034] The present application embodiment provides a method for sorting search attribute information, which is applied to electronic devices, referring to Figure 1 As shown, the method comprises the following steps:

[0035] Step 101: Perform online recall processing based on the received search terms of the product to obtain multiple recall attribute information corresponding to the search terms.

[0036] In an embodiment of the present application, the search word can be input by the user through an input device of the electronic device, and the electronic device receives the search word for the product input by the user; the search word can also be input by the user through an input device connected to the electronic device, and the electronic device receives the search word for the product input by the user.

[0037] In the embodiments of the present application, online recall processing based on search terms can be understood as online recall of products associated with the search terms based on an online recall model; online recall processing based on search terms can also be understood as online recall of products associated with the search terms based on acquired offline files, where offline files can be understood as files containing all product information stored in the memory of the electronic device or in a server connected to the electronic device.

[0038] In the embodiment of the present application, the attribute information includes attributes and attribute values; the attributes include but are not limited to the brand of the product, the color of the product, the description of the product, etc., and the attribute values ​​are the content corresponding to the attributes, and the attribute values ​​include: the specific brand of the product, the specific color of the product, the specific content and number of words describing the product, etc. For example, taking the product as a mobile phone, the attributes in the attribute information include the memory of the mobile phone, the color of the mobile phone, and the description of the mobile phone, and the attribute values ​​corresponding to each attribute, that is, the size of the mobile phone memory includes 256G, 128G and 64G, etc., and the color of the mobile phone includes black, white and gold, etc.

[0039] In the embodiment of the present application, the recalled attribute information is the attribute information corresponding to the product whose association degree with the search term satisfies the association degree condition.

[0040] In an embodiment of the present application, after receiving a search term for a product, the electronic device obtains multiple types of products associated with the search term based on the search term, and performs online recall processing on at least one product whose degree of association with the search term meets the degree of association condition. Further, the electronic device summarizes multiple recall attribute information of at least one product.

[0041] In some embodiments, after receiving a search term for a product, the electronic device obtains multiple types of products associated with the search term based on the search term, and performs online recall processing on all products whose degree of association with the search term does not meet the degree of association condition among the multiple types of products. Further, the electronic device summarizes multiple recall attribute information of the multiple types of products.

[0042] In some embodiments, the electronic device obtains a default order of attributes and attribute values ​​in the recalled attribute information of at least one product.

[0043] In some embodiments, the electronic device obtains a default classification identifier for attributes and attribute values ​​in the product's recall attribute information from a hierarchical category system (category propertyValue, CPV) of all commodities, and determines a default ordering of the attributes and attribute values ​​in the recall attribute information based on the classification identifier.

[0044] In one possible application scenario, first, the user enters the search term "mobile phone" in the search box of the user interface, and the electronic device obtains a product list based on the search term, such as mobile phones, mobile phone cases and other related products. Among them, the attribute information of the mobile phone includes but is not limited to the size of the mobile phone, the memory of the mobile phone, etc., and the attributes of the mobile phone case include but are not limited to transparency, material, etc. Figure 2 As shown, Figure 2 Shown is a recall logic diagram for multiple recalled attributes. In one case, the electronic device determines that the products in the product list that have a high correlation classification with the search term are mobile phones, obtains a list of mobile phones containing mobile phones for display, and summarizes the attributes and attribute values ​​that are highly correlated with the search term "mobile phone" among all the attributes of all mobile phones as recall attribute information, such as recalling the attributes of "mobile phone", namely mobile phone size and mobile phone memory. Based on the hierarchical category system CPV, determine the default sorting of each attribute and the default sorting of the attribute values ​​corresponding to each attribute, thereby obtaining multiple recalled attribute information after summary. In another case, the electronic device determines that the products in the product display list do not have a high correlation classification with the search term, and summarizes the attributes and attribute values ​​of all the products in the product display list.

[0045] Step 102: Determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file.

[0046] The reference attribute information is candidate attribute information selected for display, and the posterior probability of the reference attribute information is within a posterior probability threshold range.

[0047] The reference attribute information is partially or completely different from the multiple recalled attribute information.

[0048] In the embodiment of the present application, the candidate attribute information is the attribute information that satisfies the association condition with the search term and is screened out by the electronic device from the online file.

[0049] In the embodiment of the present application, the reference attribute information is the candidate attribute information whose posterior probability of the candidate attribute information selected for display is within the posterior probability threshold range. The reference attribute information is used to adjust the sorting result of the sorted multiple recalled attribute information.

[0050] In the embodiment of the present application, the posterior probability of the reference attribute information is the probability that the reference attribute information is used for display under the selected condition.

[0051] In an embodiment of the present application, attribute information that meets the association conditions can be understood as attribute information that meets the user's search intention; attribute information that meets the association conditions can also be understood as attribute information whose similarity between product words and attribute names is greater than a similarity threshold based on an online or offline corpus such as an article; attribute information that meets the association conditions can also be understood as the similarity between product words and attribute names determined based on an online or offline corpus such as an article, and based on all similarities, all attributes are sorted to determine the attribute information sorted before a preset position.

[0052] In an embodiment of the present application, the electronic device performs online recall processing based on a received search term for a product, and when multiple recall attribute information corresponding to the search term is obtained, the candidate attribute information selected for display is determined to be reference attribute information from the candidate attribute information that meets the association conditions with the search term screened out from the online file, and the posterior probability of the reference attribute information is within the posterior probability threshold range, thereby allowing the electronic device to re-sort the reference attribute information and the sorted multiple recall attribute information.

[0053] It should be noted that the electronic device can obtain the posterior probability of the candidate attribute information being selected for display in the attribute screening module through an exploration-exploitation (EE) algorithm. Here, the exploration-exploitation algorithm can be a Thompson sampling algorithm.

[0054] Step 103: Sort the reference attribute information and the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information that meets the sorting condition and corresponds to the search term for display.

[0055] In the embodiment of the present application, the multiple ordered target attribute information of the sorting condition can be understood as the attribute information whose sorting position is determined before the preset position, such as the top ten positions, after the reference attribute information and the sorted multiple recall attributes are re-sorted.

[0056] In an embodiment of the present application, the plurality of ordered target attribute information may include only a plurality of recalled attribute information; the plurality of ordered target attribute information may also include both recalled attribute information and reference attribute information.

[0057] In an embodiment of the present application, the electronic device determines the reference attribute information from the candidate attribute information that meets the association conditions with the search term screened out from the online file, and then sorts the reference attribute information with the sorted multiple recalled attribute information to obtain multiple ordered target attribute information that meets the sorting conditions and corresponds to the search term for display.

[0058] In an embodiment of the present application, a method for sorting search attribute information is provided. An online recall process is performed based on a received search term of a product to obtain a plurality of recalled attribute information corresponding to the search term; reference attribute information is determined from candidate attribute information that meets the association condition with the search term screened out from an online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially or completely different from the plurality of recalled attribute information; the reference attribute information is sorted with the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information that meets the sorting condition and corresponds to the search term for display; in this way, the problem that at least the offline sorting result obtained by a linear model is relatively stable and there is a certain delay in updating the sorted attributes to the search engine in the related art is solved, and the reference attribute information whose posterior probability is within the posterior probability threshold range is added to the sorted plurality of recalled attribute information, thereby avoiding the Matthew effect caused by attribute screening, and providing accurate and rich attribute sorting results for the attribute information in the screening attribute module, maximizing the click-through rate and conversion rate of the screening, thereby shortening the user's browsing and purchasing path.

[0059] The present application embodiment provides a method for sorting search attribute information, which is applied to electronic devices, referring to Figure 3 As shown, the method comprises the following steps:

[0060] Step 201: Perform online recall processing based on the received search terms of the product to obtain multiple recall attribute information corresponding to the search terms.

[0061] Step 202: Determine reference attribute information from candidate attribute information that meets the association condition with the search term and is screened out from the online file.

[0062] The reference attribute information is candidate attribute information selected for display, and the posterior probability of the reference attribute information is within a posterior probability threshold range.

[0063] The reference attribute information is partially or completely different from the multiple recalled attribute information.

[0064] In the present application, refer to Figure 4As shown, step 202 determines reference attribute information from candidate attribute information that meets the association condition with the search term screened out from the online file, which can be implemented by the following steps:

[0065] Step 2021: Determine the product terms included in the search terms.

[0066] In the embodiment of the present application, the electronic device performs word segmentation on the search term, that is, the search term is divided into multiple word segmentation terms to determine the product terms contained in the search term. Here, it should be noted that the word segmentation and the part-of-speech information corresponding to the word segmentation can be used in modules such as term importance analysis and intent recognition, so that the attributes can be effectively screened. The current word segmentation technology is relatively mature, and this application does not provide a specific description of this.

[0067] Step 2022: Filter out from the first file first product attribute information whose similarity with the product attribute information corresponding to the product word meets the first similarity threshold; and / or filter out from the second file second product attribute information whose similarity with the product word meets the second similarity threshold.

[0068] The candidate attribute information includes first product attribute information and second product attribute information, and the first file and the second file are different.

[0069] In the embodiment of the present application, the first file can be understood as an online file formed by the search term input by the user. The second file can be understood as an online corpus such as a file corresponding to an article.

[0070] In the embodiment of the present application, the first product attribute information is attribute information whose similarity with the product attribute to be expressed by the product word contained in the search word meets the first similarity threshold; the first product attribute information can also be understood as the product attribute to be expressed based on the search word. Exemplarily, the search word entered by the user is "single-slot basin", and for the product word "basin", the shape attribute of the basin is relatively important, then it is determined that the similarity between the basin and the shape attribute meets the first similarity threshold, and the shape attribute is determined to be the first product attribute information.

[0071] In the embodiment of the present application, the second product attribute information is attribute information whose similarity between the product word and the product attribute information meets the second similarity threshold.

[0072] In one application scenario, since both product words and attribute information exist in the corpus, and both product words and attribute information are text, first, the electronic device uses the screening feedback data fine tune BERT model to determine the attribute information associated with the product words. Secondly, the electronic device inputs the product words and attribute information into the segment-pair model to obtain the semantic similarity between the product words and attribute information output by the segment-pair model, and determines that the attribute information whose similarity is greater than a certain threshold or whose ranking position of the similarity of the attribute information under the product word is before the preset ranking position is the second product attribute information of the product word.

[0073] In the embodiment of the present application, the electronic device has too many attributes to explore, and the click rate of clicking on the attributes in the attribute screening module is only 3%. It is impossible to detect all attributes when the traffic is limited. Considering the model convergence speed and screening efficiency, only the determined candidate attribute information is explored.

[0074] Step 2023: Determine, from the candidate attribute information, the candidate attribute information whose posterior probability is within the posterior probability threshold range as the reference attribute information.

[0075] Step 203: Obtain the posterior probability that the reference attribute information is selected for display.

[0076] In the embodiment of the present application, the posterior probability of the reference attribute information is the probability that the reference attribute information is used for display under the selected condition.

[0077] In an embodiment of the present application, the electronic device determines a search term from the acquired search log, and based on the attribute information corresponding to the search term and the search behavior information corresponding to the search term, determines the posterior probability that the reference attribute information associated with the search term is selected for display through the EE algorithm.

[0078] Step 204: based on the posterior probability, sort the reference attribute information and the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information that meets the sorting condition and corresponds to the search term for display.

[0079] In the embodiment of the present application, after the electronic device obtains the posterior probability that the reference attribute information is selected for display, the reference attribute information and the sorted multiple recalled attribute information are sorted based on the posterior probability of the reference attribute information to obtain multiple ordered target attribute information corresponding to the search term for display that meets the sorting condition. Among them, the attributes displayed in the attribute screening module are multiple ordered target attribute information.

[0080] In some embodiments, the electronic device determines that the posterior probability of the reference attribute information, such as 0.85, is within the first probability threshold range of 0.8 to 0.9, and obtains the position information corresponding to the first probability threshold range through a linear function, such as the fifth position in the sorted multiple recalled attribute information, and moves the fifth position and other recalled attribute information after the fifth position in the sorted multiple recalled attribute information backward by one position, intersperses the reference attribute information in the fifth position, and obtains multiple ordered target attribute information in the top ten positions after re-sorting.

[0081] In other embodiments, the electronic device determines that the posterior probability of the reference attribute information, such as 0.75, is within the second probability threshold range of 0.7 to 0.8, and obtains the position information corresponding to the second probability threshold range through a linear function, such as the eighth position in the sorted multiple recalled attribute information, and moves the eighth position and other recalled attribute information after the eighth position in the sorted multiple recalled attribute information backward by one position, intersperses the reference attribute information at the eighth position, and obtains multiple ordered target attribute information in the top ten positions after re-sorting.

[0082] In some further embodiments, the electronic device determines that the posterior probability of the reference attribute information, such as 0.1, is within a third probability threshold range, and obtains position information corresponding to the third probability threshold range, such as the last position in the sorted multiple recalled attribute information, through a linear function, and intersperses the reference attribute information with the last position in the multiple recalled attribute information to obtain multiple ordered target attribute information in the top ten positions after re-sorting.

[0083] In the present application, refer to Figure 3 As shown, after step 204 sorts the reference attribute information and the sorted multiple recalled attribute information based on the posterior probability to obtain multiple ordered target attribute information corresponding to the search term for display and satisfying the sorting condition, the following steps may be performed:

[0084] Step 205: If the target attribute information includes reference attribute information, obtain a weight coefficient corresponding to the posterior probability when the reference attribute information is displayed.

[0085] In an embodiment of the present application, if the target attribute information includes reference attribute information, the electronic device obtains the reference attribute information through an exploration-exploitation (EE) algorithm and displays it in the attribute screening module, and the weight coefficient corresponding to the posterior probability.

[0086] Here, the detection and display algorithm can be a Thompson sampling algorithm. The specific implementation method of the Thompson sampling algorithm is shown below.

[0087]

[0088] In the above-mentioned Thompson sampling algorithm, alpah and beta are prior parameters. Si is the success count of clicking on the reference attribute information displayed in the filtering attribute module, and Fi is the failure count of not clicking on the reference attribute information displayed in the filtering attribute module. T is the number of times the reference attribute information is explored, and K is the number of reference attribute information. Bidirectional Encoder Representation from Transformers is the beta distribution function. r is the weight coefficient, which is 1 for the binomial distribution, and based on the weight coefficient, the success count Si or the failure count Fi is processed. Exemplarily, if the user clicks on the reference attribute information displayed in the filtering attribute module, the weight coefficient is 1, and the corresponding success count Si is increased by 1. If the user does not click on the reference attribute information displayed in the filtering attribute module, the corresponding failure count Fi is increased by 1.

[0089] The above-mentioned Beta is a pre-trained language model. First, a large amount of unsupervised corpus is used for language model pre-training (Pre-training), and then a small amount of annotated corpus is used for fine-tuning (Fine-tuning) to complete specific natural language processing tasks. Among them, natural language processing tasks include but are not limited to classification, sequence labeling, sentence relationship judgment and machine reading comprehension.

[0090] Step 206: Update the posterior probability of the reference attribute information being selected for display based on the weight coefficient.

[0091] In an embodiment of the present application, the electronic device determines the success count of the reference attribute information displayed in the filtering attribute module, or determines the count of the second reference attribute information displayed in the filtering attribute module based on the weight coefficient, and updates the posterior probability that the reference attribute information is selected for display based on the success count or the failure count.

[0092] Step 207: Based on the updated posterior probability, adjust the order of the reference attribute information in the plurality of ordered target attribute information.

[0093] In the embodiment of the present application, the order of the reference attribute information in the plurality of ordered target attribute information is adjusted based on the updated posterior probability.

[0094] In a feasible application scenario, refer to Figure 5 As shown, Figure 5What is shown is a structural diagram of multiple ordered attribute information displayed for reference attribute information and user feedback behavior data. The electronic device obtains candidate attribute information based on the search term, and stores the candidate attribute information in the feature service (FS), and reads the candidate attribute information and the user feedback behavior data output in real time by Flink from the FS online. The electronic device uses the EE algorithm to decide the candidate attribute information with a high posterior probability of being selected for display as the reference attribute information, and determines the reference attribute information based on the posterior probability of the reference attribute information, and the location information displayed in the attribute screening module. When the user sees the displayed reference attribute information, feedback behavior data such as clicks or views will be generated, and then these feedback behavior data will flow back to the system to help the system decide the reference attribute information that should be detected next.

[0095] As can be seen from the above, in the embodiment of the present application, first, an online recall process is performed based on the received search terms of the product to obtain multiple recall attribute information corresponding to the search terms. Secondly, the electronic device determines the product terms contained in the search terms, and screens out from the first file the first product attribute information whose similarity with the product attribute information corresponding to the product terms meets the first similarity threshold; and / or, screens out from the second file the second product attribute information whose similarity with the product terms meets the second similarity threshold. And determine the reference attribute information from the candidate attribute information, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range. Thirdly, the electronic device obtains the posterior probability that the reference attribute information is selected for display, and based on the posterior probability, sorts the reference attribute information with the sorted multiple recall attribute information to obtain multiple ordered target attribute information that meet the sorting conditions corresponding to the search terms for display. Finally, the electronic device uses the EE algorithm to obtain the weight coefficient corresponding to the posterior probability when the reference attribute information is displayed, updates the posterior probability of the reference attribute information being selected for display based on the weight coefficient, and adjusts the order of the reference attribute information in the multiple ordered target attribute information based on the updated posterior probability. In this way, the electronic device uses the EE algorithm to add the reference attribute information whose posterior probability is within the posterior probability threshold range to the multiple recalled attribute information after sorting, reducing the Matthew effect, and at the same time providing accurate and rich attribute sorting results for the attribute information in the screening attribute module, maximizing the click-through rate and conversion rate of the screening, thereby shortening the user's browsing and purchasing path.

[0096] The present application embodiment provides a method for sorting search attribute information, which is applied to electronic devices, referring to Figure 6 As shown, the method comprises the following steps:

[0097] Step 301: Perform online recall processing based on the received search terms of the product to obtain multiple recall attribute information corresponding to the search terms.

[0098] Step 302: input the search term and the multiple recalled attribute information into the trained attribute ranking model to obtain the sorted multiple recalled attribute information output by the attribute ranking model.

[0099] In an embodiment of the present application, the attribute ranking model can be a model trained based on sample search terms in the user's historical search log, attribute features corresponding to the sample search terms, and the user's search behavior information.

[0100] In the present application, refer to Figure 7 As shown, the attribute ranking model in step 302 can be obtained through the following steps:

[0101] Step 3021: Obtain sample search terms from historical search logs.

[0102] In an embodiment of the present application, the historical search log refers to a collection of a series of actions performed continuously by the user within an interval of a predetermined time period. Each log in the historical search log includes, but is not limited to, search user cookies, search behavior information of the user using the search engine, the time when the search behavior occurred, the location where the search behavior occurred, and the search words corresponding to the search behavior. Here, a series of actions includes the input historical search words and the search behavior information for each historical search word, such as search behavior, click behavior on the attributes of the attribute screening module, click behavior on the search results, page turning behavior, and page closing behavior. Due to the massive content of the historical search log, only a certain number of logs within a predetermined time period before the current moment are extracted in an embodiment of the present application.

[0103] Step 3022: Generalize the sample search terms based on the historical search logs to obtain multi-dimensional aggregate features corresponding to the sample search terms.

[0104] Among them, the multi-dimensional aggregated features are composed of features of different dimensions and attribute information corresponding to sample search terms.

[0105] In some embodiments, features of different dimensions include: search terms, product terms, and categories corresponding to the recalled attribute information.

[0106] In the embodiment of the present application, features of different dimensions are determined based on sample search terms.

[0107] In an embodiment of the present application, the electronic device generalizes the sample search terms based on the historical search logs to obtain multi-dimensional aggregate features corresponding to the sample search terms, which are composed of features of different dimensions and attribute information corresponding to the sample search terms.

[0108] In an embodiment of the present application, first, the electronic device obtains a historical search log between the current moment and the time dimension within a preset time period. Secondly, the electronic device obtains the sample search term input by the user, as well as all attribute information corresponding to the sample search term, the product term included in the sample search term, and the category to which the sample search term belongs from the historical search log. Then, the electronic device generalizes the sample search term based on obtaining the sample search term input by the user, all attribute information corresponding to the sample search term, the product term included in the sample search term, and the category to which the sample search term belongs, and obtains a multi-dimensional aggregation feature corresponding to the sample search term, wherein the multi-dimensional aggregation feature includes an aggregation feature composed of the sample search term and all attribute information corresponding to the sample search term, an aggregation feature composed of the product term included in the sample search term and all attribute information corresponding to the sample search term, and an aggregation feature composed of the category to which the sample search term belongs and all attribute information corresponding to the sample search term. In this way, by multi-dimensionally modeling the relationship between search terms and the attributes corresponding to the search terms, when the search term currently input by the user is a new term, that is, a term that no user has searched for before, the current search term can be generalized to an aggregate feature of at least one dimension among the multi-dimensional aggregate features, thereby improving the generalization ability of the electronic device for search terms.

[0109] In an achievable application scenario, when the user inputs the search term "ultra high-definition television", the electronic device does not find the search term "ultra high-definition television" from the historical search log, but finds the sample search term "64-inch television". Since no user has searched for the search term "ultra high-definition television", the feedback data of the attribute information of the search term "ultra high-definition television" is missing. At this time, the electronic device can use the aggregated features composed of the product term "television" contained in the search term and all the attribute information corresponding to the sample search term to feed back the attribute information of all televisions corresponding to the sample search term "64-inch television" to the attribute information corresponding to the search term "ultra high-definition television" and display it. In this way, the relationship between the search term and the attribute corresponding to the search term is modeled in multiple dimensions, and the generalization ability of the electronic device for the search term is improved.

[0110] Step 3023: Train an attribute ranking model based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms.

[0111] In the embodiment of the present application, the search behavior information is the behavior information of the attributes at different positions in the displayed attribute filtering module when the user uses the sample search term, such as the behavior of selecting an attribute at a certain position and clicking to filter it.

[0112] In the embodiment of the present application, the electronic device generates training samples based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms. The extreme gradient boosting model (eXtreme GradientBoosting, xgboost) is used to perform model training processing on the training samples to generate an attribute ranking model.

[0113] In an achievable application scenario, the electronic device generates training samples based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms. The combination here is that the training samples are composed of features and target labels. In one case, if the user clicks to filter the attributes in the filter attribute module based on the sample search terms, the target label of the sample is 1, which is a positive sample, and the target label of the attribute that is not clicked to filter is 0, which is a negative sample. In another case, if the sample search term entered by the user contains attributes, such as "256G mobile phone", then the mobile phone memory of the attribute under the mobile phone is a positive sample.

[0114] In some embodiments, the electronic device generates a test sample based on aggregated features and historical search behavior information of historical search terms; inputs the test sample into an initial attribute ranking model to obtain an initial attribute ranking output by the initial attribute ranking model; based on the initial attribute ranking and the acquired real attribute ranking, obtains a receiver operating characteristic (ROC) curve, i.e., an attribute ranking result; if the area under the ROC curve (Area Under the Curve, AUC) value such as 0.73% meets the AUC threshold range, the initial attribute ranking model is determined to be an attribute ranking model, wherein the AUC value is used to evaluate the ranking result of the initial attribute ranking model.

[0115] In an achievable application scenario, for example, when the user inputs "mobile hard disk", the electronic device inputs the mobile hard disk and the multiple recalled attribute information corresponding to the mobile hard disk into the attribute sorting model, and the attribute sorting model outputs that the importance of the "capacity" attribute is greater than the "warranty" attribute. The electronic device ranks the "capacity" attribute before the "warranty" attribute. For another example, when the user inputs "lipstick", the electronic device inputs the lipstick and the multiple recalled attribute information corresponding to the lipstick into the attribute sorting model, and the attribute sorting model outputs that the importance of the "color" attribute is greater than the "makeup effect" attribute. The electronic device ranks the "color" attribute before the "makeup effect" attribute.

[0116] In some embodiments, the electronic device may calculate the AUC value based on all samples to obtain the Global AUC which is the overall ranking effect reflected by all samples.

[0117] In some embodiments, the electronic device may also calculate an AUC value based on a search sample of the user to obtain an average ranking effect session AUC reflected by each search sample.

[0118] As can be seen from the above, in the embodiment of the present application, the electronic device obtains an attribute ranking model based on the multi-dimensional aggregation features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms, which is an attribute ranking that conforms to the user behavior. In this way, when the user enters a search term, the electronic device inputs the search term and the multiple recalled attribute information corresponding to the search term into the trained attribute ranking model, and obtains the sorted multiple recalled attribute information output by the attribute ranking model, so as to provide accurate and rich attribute ranking results, maximize the click-through rate and conversion rate of the screening, and thus shorten the user's browsing and purchasing path.

[0119] Step 303: Determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file.

[0120] The reference attribute information is candidate attribute information selected for display, and the posterior probability of the reference attribute information is within a posterior probability threshold range.

[0121] The reference attribute information is partially or completely different from the multiple recalled attribute information.

[0122] Step 304: Sort the reference attribute information and the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information that meets the sorting condition and corresponds to the search term for display.

[0123] As can be seen from the above, in the embodiment of the present application, the electronic device uses multi-dimensional modeling to establish the relationship between the search term and the attribute corresponding to the search term. When the search term currently input by the user is a new term, the current search term can still be generalized to the aggregated features of at least one dimension in the multi-dimensional aggregated features, thereby improving the generalization ability of the electronic device to the search term. At the same time, when the user inputs the search term, the electronic device inputs the search term and the multiple recalled attribute information corresponding to the search term into the trained attribute sorting model, and obtains the sorted multiple recalled attribute information output by the attribute sorting model. In this way, accurate and rich attribute sorting results are provided, which maximizes the click-through rate and conversion rate of the screening and shortens the user's browsing and purchasing path.

[0124] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0125] The present application provides a device for sorting search attribute information. The device for sorting search attribute information can be used to implement Figure 1, Figure 3-Figure 4 , Figure 6-Figure 7 A method for sorting search attribute information provided by a corresponding embodiment is as follows: Figure 8 As shown, the sorting device 8 for searching attribute information includes:

[0126] The first processing module 81 is used to perform online recall processing based on the received search terms of the product, and obtain multiple recall attribute information corresponding to the search terms;

[0127] The second processing module 82 is used to determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information;

[0128] The sorting module 83 is used to sort the reference attribute information and the sorted plurality of recalled attribute information.

[0129] In other embodiments of the present application, the determination module is used to determine the product words included in the search words; the second processing module 82 is also used to filter out from the first file the first product attribute information whose similarity with the product attribute information corresponding to the product words meets the first similarity threshold; and / or, filter out from the second file the second product attribute information whose similarity with the product words meets the second similarity threshold; wherein the candidate attribute information includes the first product attribute information and the second product attribute information, and the first file and the second file are different; the determination module is also used to determine from the candidate attribute information, the candidate attribute information whose posterior probability of the candidate attribute information is within the posterior probability threshold range as the reference attribute information.

[0130] In other embodiments of the present application, the acquisition module is used to obtain the posterior probability that the reference attribute information is selected for display; the sorting module 83 is also used to sort the reference attribute information and the sorted multiple recall attribute information based on the posterior probability, to obtain multiple ordered target attribute information that meets the sorting conditions and corresponds to the search terms for display.

[0131] In other embodiments of the present application, the acquisition module is used to obtain the weight coefficient corresponding to the posterior probability of the reference attribute information when it is displayed if the target attribute information includes reference attribute information; the second processing module 82 is also used to update the posterior probability of the reference attribute information being selected for display based on the weight coefficient; the sorting module 83 is also used to adjust the sorting of the reference attribute information in multiple ordered target attribute information based on the updated posterior probability.

[0132] In other embodiments of the present application, the first processing module 81 is further used to input the search term and the multiple recalled attribute information into a trained attribute ranking model to obtain the sorted multiple recalled attribute information output by the attribute ranking model.

[0133] In other embodiments of the present application, the acquisition module is also used to obtain sample search terms from historical search logs; the first processing module 81 is also used to generalize the sample search terms based on the historical search logs to obtain multi-dimensional aggregate features corresponding to the sample search terms; wherein the multi-dimensional aggregate features are composed of features of different dimensions and attribute information corresponding to the sample search terms; based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms, the attribute ranking model is trained.

[0134] In other embodiments of the present application, features of different dimensions include: search terms, product terms, and categories corresponding to recall attribute information.

[0135] The present invention provides an electronic device which can be used in Figure 1 , Figure 3-Figure 4 , Figure 6-Figure 7 In a method for sorting search attribute information provided by a corresponding embodiment, referring to Fig. 9 As shown, the electronic device ( Fig. 9 Electronic devices 9 in the Figure 8 The search attribute information sorting device 8) comprises: a processor 91, a memory 92 and a communication bus 93, wherein:

[0136] The communication bus 93 is used to realize the communication connection between the processor 91 and the memory 92;

[0137] The processor 91 is used to execute the sorting program of the search attribute information stored in the memory 92 to implement the following steps:

[0138] Performing online recall processing based on the received search terms of the product, and obtaining multiple recall attribute information corresponding to the search terms;

[0139] Determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information;

[0140] The reference attribute information and the sorted multiple recalled attribute information are sorted to obtain multiple ordered target attribute information that meet the sorting conditions and correspond to the search terms for display.

[0141] In other embodiments of the present application, the processor 91 is used to execute the sorting processing program of the search attribute information stored in the memory 92 to implement the following steps:

[0142] Determine product words included in the search words; select from the first file first product attribute information whose similarity with the product attribute information corresponding to the product words meets the first similarity threshold; and / or select from the second file second product attribute information whose similarity with the product words meets the second similarity threshold; wherein the candidate attribute information includes the first product attribute information and the second product attribute information, and the first file and the second file are different; from the candidate attribute information, determine the candidate attribute information whose posterior probability is within the posterior probability threshold as the reference attribute information.

[0143] In other embodiments of the present application, the processor 91 is used to execute the sorting processing program of the search attribute information stored in the memory 92 to implement the following steps:

[0144] Obtain a posterior probability that the reference attribute information is selected for display; based on the posterior probability, sort the reference attribute information and the sorted multiple recalled attribute information to obtain multiple ordered target attribute information that meets the sorting conditions and corresponds to the search term for display.

[0145] In other embodiments of the present application, the processor 91 is used to execute the sorting processing program of the search attribute information stored in the memory 92 to implement the following steps:

[0146] If the target attribute information includes reference attribute information, obtain the weight coefficient corresponding to the posterior probability when the reference attribute information is displayed; update the posterior probability that the reference attribute information is selected for display based on the weight coefficient; and adjust the order of the reference attribute information in the multiple ordered target attribute information based on the updated posterior probability.

[0147] In other embodiments of the present application, the processor 91 is used to execute the sorting processing program of the search attribute information stored in the memory 92 to implement the following steps:

[0148] The search term and the multiple recalled attribute information are input into the trained attribute ranking model to obtain the sorted multiple recalled attribute information output by the attribute ranking model.

[0149] In other embodiments of the present application, the processor 91 is used to execute the sorting processing program of the search attribute information stored in the memory 92 to implement the following steps:

[0150] Sample search terms are obtained from historical search logs; the sample search terms are generalized based on the historical search logs to obtain multi-dimensional aggregate features corresponding to the sample search terms; wherein the multi-dimensional aggregate features are composed of features of different dimensions and attribute information corresponding to the sample search terms; based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms, an attribute ranking model is trained.

[0151] In other embodiments of the present application, features of different dimensions include: search terms, product terms, and categories corresponding to recall attribute information.

[0152] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0153] Based on the foregoing embodiments, an embodiment of the present application provides a computer storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the following steps:

[0154] Performing online recall processing based on the received search terms of the product, and obtaining multiple recall attribute information corresponding to the search terms;

[0155] Determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information;

[0156] The reference attribute information and the sorted multiple recalled attribute information are sorted to obtain multiple ordered target attribute information that meet the sorting conditions and correspond to the search terms for display.

[0157] In other embodiments of the present application, the one or more programs may be executed by one or more processors, and the following steps may also be implemented:

[0158] Determine product words included in the search words; select from the first file first product attribute information whose similarity with the product attribute information corresponding to the product words meets the first similarity threshold; and / or select from the second file second product attribute information whose similarity with the product words meets the second similarity threshold; wherein the candidate attribute information includes the first product attribute information and the second product attribute information, and the first file and the second file are different; from the candidate attribute information, determine the candidate attribute information whose posterior probability is within the posterior probability threshold as the reference attribute information.

[0159] In other embodiments of the present application, the one or more programs may be executed by one or more processors, and the following steps may also be implemented:

[0160] Obtain a posterior probability that the reference attribute information is selected for display; based on the posterior probability, sort the reference attribute information and the sorted multiple recalled attribute information to obtain multiple ordered target attribute information that meets the sorting conditions and corresponds to the search term for display.

[0161] In other embodiments of the present application, the one or more programs may be executed by one or more processors, and the following steps may also be implemented:

[0162] If the target attribute information includes reference attribute information, obtain the weight coefficient corresponding to the posterior probability when the reference attribute information is displayed; update the posterior probability that the reference attribute information is selected for display based on the weight coefficient; and adjust the order of the reference attribute information in the multiple ordered target attribute information based on the updated posterior probability.

[0163] In other embodiments of the present application, the one or more programs may be executed by one or more processors, and the following steps may also be implemented:

[0164] The search term and the multiple recalled attribute information are input into the trained attribute ranking model to obtain the sorted multiple recalled attribute information output by the attribute ranking model.

[0165] In other embodiments of the present application, the one or more programs may be executed by one or more processors, and the following steps may also be implemented:

[0166] Sample search terms are obtained from historical search logs; the sample search terms are generalized based on the historical search logs to obtain multi-dimensional aggregate features corresponding to the sample search terms; wherein the multi-dimensional aggregate features are composed of features of different dimensions and attribute information corresponding to the sample search terms; based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms, an attribute ranking model is trained.

[0167] In other embodiments of the present application, features of different dimensions include: search terms, product terms, and categories corresponding to recall attribute information.

[0168] It should be noted that the above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM) and other memories; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0169] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0170] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0171] In addition, all functional units in the embodiments of the present application can be integrated into one processing module, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units. A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium, which, when executed, executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memory (ROM), random access memory (RAM), disks or optical disks, and other media that can store program codes.

[0172] The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0173] The features disclosed in several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0174] The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0175] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for sorting search attribute information, characterized in that: The method comprises: Performing online recall processing based on the received search terms of the product to obtain multiple recall attribute information corresponding to the search terms; Determining reference attribute information from candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information; Sorting the reference attribute information and the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information corresponding to the search term for display and satisfying a sorting condition; The step of sorting the reference attribute information and the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information corresponding to the search term for display and satisfying a sorting condition includes: Obtaining the posterior probability that the reference attribute information is selected for display; Based on the posterior probability, the reference attribute information and the sorted plurality of recalled attribute information are sorted to obtain a plurality of ordered target attribute information corresponding to the search term for display and meeting the sorting condition.

2. The method according to claim 1, characterized in that The online file includes a first file and a second file, and determining reference attribute information from candidate attribute information that meets the association condition between the search term and the online file, including: Determine product terms included in the search terms; Filtering out, from the first file, first product attribute information whose similarity with the product attribute information corresponding to the product word meets a first similarity threshold; and / or, Filtering out, from the second file, second product attribute information whose similarity to the product word meets a second similarity threshold; wherein the candidate attribute information includes first product attribute information and second product attribute information, and the first file and the second file are different; From the candidate attribute information, determine the candidate attribute information whose posterior probability is within the posterior probability threshold range as the reference attribute information.

3. The method according to claim 1, characterized in that If the target attribute information includes the reference attribute information, after the reference attribute information and the sorted plurality of recalled attribute information are sorted based on the posterior probability to obtain a plurality of ordered target attribute information corresponding to the search term for display and satisfying the sorting condition, the method includes: Obtaining a weight coefficient corresponding to the posterior probability when the reference attribute information is displayed; updating the posterior probability that the reference attribute information is selected for display based on the weight coefficient; Based on the updated posterior probability, the order of the reference attribute information in the plurality of ordered target attribute information is adjusted.

4. The method according to claim 1, characterized in that After performing online recall processing based on the received product search terms and obtaining a plurality of recall attribute information corresponding to the search terms, the method includes: The search term and the plurality of recalled attribute information are input into a trained attribute ranking model to obtain the sorted plurality of recalled attribute information output by the attribute ranking model.

5. The method according to claim 4, characterized in that The method further comprises: Get sample search terms from historical search logs; Generalizing the sample search terms based on the historical search logs to obtain multi-dimensional aggregate features corresponding to the sample search terms; wherein the multi-dimensional aggregate features are composed of features of different dimensions and attribute information corresponding to the sample search terms; The attribute ranking model is trained based on the multi-dimensional aggregate features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms.

6. The method according to claim 5, characterized in that The features of different dimensions include: search terms, product terms, and categories corresponding to the recalled attribute information.

7. A device for sorting search attribute information, characterized in that: The device comprises: A first processing module, configured to perform online recall processing based on the received search terms of the product, and obtain multiple recall attribute information corresponding to the search terms; A second processing module is used to determine reference attribute information from the candidate attribute information that meets the association condition with the search term screened out from the online file, wherein the reference attribute information is the candidate attribute information selected for display, and the posterior probability of the reference attribute information is within the posterior probability threshold range, and the reference attribute information is partially different or completely different from the multiple recalled attribute information; A sorting module, used to sort the reference attribute information and the sorted plurality of recalled attribute information to obtain a plurality of ordered target attribute information corresponding to the search term for display and meeting the sorting condition; The device also includes: An acquisition module, used for acquiring the posterior probability that the reference attribute information is selected for display; The sorting module is further used to sort the reference attribute information and the sorted multiple recalled attribute information based on the posterior probability, so as to obtain multiple ordered target attribute information corresponding to the search term for display and meeting the sorting condition.

8. The device according to claim 7, characterized in that The device also includes: A determination module, used to determine product words included in the search words; The second processing module is further configured to filter out, from the first file included in the online file, first product attribute information whose similarity with the product attribute information corresponding to the product word meets a first similarity threshold; and / or, The second processing module is further used to filter out, from the second file included in the online file, second product attribute information whose similarity with the product word meets a second similarity threshold; wherein the candidate attribute information includes first product attribute information and second product attribute information, and the first file and the second file are different; The determination module is further configured to determine, from the candidate attribute information, candidate attribute information whose posterior probability of the candidate attribute information is within the posterior probability threshold range as the reference attribute information.

9. The device according to claim 7, characterized in that The acquisition module is further used to acquire a weight coefficient corresponding to the posterior probability when the reference attribute information is displayed; The second processing module is further used to update the posterior probability of the reference attribute information being selected for display based on the weight coefficient; The sorting module is further used to adjust the sorting of the reference attribute information in the plurality of ordered target attribute information based on the updated posterior probability.

10. The device according to claim 7, characterized in that The first processing module is further used to input the search term and the multiple recalled attribute information into a trained attribute ranking model to obtain the sorted multiple recalled attribute information output by the attribute ranking model.

11. The device according to claim 10, characterized in that The acquisition module is further used to acquire sample search terms from historical search logs; The first processing module is further used to perform generalization processing on the sample search term based on the historical search log to obtain multi-dimensional aggregate features corresponding to the sample search term; wherein the multi-dimensional aggregate features are composed of features of different dimensions and attribute information corresponding to the sample search term; The first processing module is further used to train the attribute ranking model based on the multi-dimensional aggregation features corresponding to the sample search terms and the search behavior information corresponding to the sample search terms.

12. The device according to claim 11, characterized in that The features of different dimensions include: search terms, product terms, and categories corresponding to the recalled attribute information.

13. An electronic device, characterized in that: The electronic device comprises: a processor, a memory and a communication bus; The communication bus is used to realize the communication connection between the processor and the memory; The processor is used to execute a sorting program of the search attribute information stored in the memory to implement the sorting method of the search attribute information as described in any one of claims 1 to 6.

14. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method for sorting search attribute information as described in any one of claims 1 to 6.

Images