Sorting method, device, equipment and storage medium

Abstract

The present invention discloses a sorting method, device, equipment and storage medium. The method comprises: obtaining user behavior information of a target user on multiple commodities, the user behavior information comprising multiple query items associated with the multiple commodities; determining the recall query items of the target user according to the similarity between the multiple query items and the target user; sorting the weights of the recall query items, and displaying the preference query items in the recall query items that meet the preset weight threshold. The embodiment of the present invention can first filter based on the query items of the recommended commodities to obtain the recall query items, and then filter the recall query items according to the weights of the recall query items, and recommend the commodities corresponding to the preference query items in the recall query items that meet the preset weight threshold to the user, thereby achieving accurate recommendation of commodities and improving user experience.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a sorting method, device, equipment and storage medium. Background Art

[0002] At present, product recommendations through recommendation systems have been widely used in industry.

[0003] Usually, before recommending products to users, the recommendation system will analyze and calculate the user's historical operation records on the products, and sort a large number of products based on the calculation results, so as to recommend products that users need. However, the number of products corresponding to the recommended products is often very large. If only the above sorting conditions are used, a large number of products will be recommended to users, and the products cannot be accurately recommended to users, resulting in poor user experience. Summary of the invention

[0004] The embodiments of the present invention provide a sorting method, apparatus, device and storage medium, which can achieve accurate recommendation of commodities.

[0005] According to one aspect of an embodiment of the present invention, a sorting method is provided, the sorting method comprising:

[0006] Obtaining user behavior information of a target user on multiple commodities, where the user behavior information includes multiple query items associated with the multiple commodities;

[0007] Determine a recall query item of the target user according to similarities between the multiple query items and the target user;

[0008] The weights of the recalled query items are sorted, and the preferred query items among the recalled query items that meet the preset weight threshold are displayed.

[0009] According to another aspect of an embodiment of the present invention, a sorting device is provided, the sorting device comprising:

[0010] An acquisition module, used to acquire user behavior information of a target user on a plurality of commodities, wherein the user behavior information includes a plurality of query items associated with the plurality of commodities;

[0011] A determination module, used to determine a recall query item of the target user according to similarities between the multiple query items and the target user;

[0012] The sorting module is used to sort the weights of the recalled query items and display the preferred query items in the recalled query items that meet the preset weight threshold.

[0013] According to another aspect of an embodiment of the present invention, an electronic device is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the sorting method described in the first aspect is implemented.

[0014] According to another aspect of the embodiments of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes instructions, and when the instructions are executed on a computer, the computer executes the sorting method described in the first aspect.

[0015] According to the sorting method, apparatus, device and storage medium in the embodiment of the present invention, the recall query items of the target user are determined by calculating the user behavior information including multiple query items associated with the target user and the similarity of the target user; and then the weights of the recall query items are sorted according to the weights of the recall query items, so as to obtain the preference query items that meet the preset weight threshold. The embodiment of the present invention can first filter based on the query items of the recommended products to obtain the recall query items, and then filter the recall query items according to the weights of the recall query items, and recommend the products corresponding to the preference query items that meet the preset weight threshold in the recall query items to the user, so as to achieve accurate recommendation of the products and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following briefly introduces the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is an example of an Internet access system architecture diagram of this specification;

[0018] Figure 2 A flowchart of a sorting method provided by an embodiment of the present invention;

[0019] Figure 3 A schematic diagram of a meta-path provided by an embodiment of the present invention;

[0020] Figure 4 A schematic diagram of a GRU-Attention mechanism provided according to an embodiment of the present invention is shown;

[0021] Figure 5 A schematic diagram of the structure of a sorting model provided according to an embodiment of the present invention is shown;

[0022] Figure 6 A schematic diagram of the structure of a recommendation system provided according to an embodiment of the present invention is shown;

[0023] Figure 7 A schematic diagram of the structure of a sorting device provided according to an embodiment of the present invention is shown;

[0024] Figure 8 A schematic diagram of a hardware structure 800 of a computing device according to an embodiment of the invention is shown. DETAILED DESCRIPTION

[0025] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and Examples. It should be understood that the specific embodiments described herein are only configured to explain the present invention and are not configured to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.

[0026] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0027] Figure 1 FIG. 1 is an example of an Internet access system architecture diagram of this specification. Figure 1 As shown, the Internet access system includes a user terminal 100, a user terminal 200, a user terminal 300 and a server 400. Figure 1Only three user terminals are used as examples. When users operate user terminal 110, user terminal 120 and user terminal 130 respectively, user behavior information will be generated on user terminal 110, user terminal 120 and user terminal 130, and the user behavior information includes multiple query items associated with multiple commodities; server 400 is used to obtain user behavior information, and determine the recall query items of the target user from the multiple query items based on the similarity between the multiple query items and the target user; then sort the weights of the recall query items, and display the preference query items in the recall query items that meet the preset weight threshold. In order to make full use of these acquired user feature information to achieve accurate recommendation of commodities, this specification provides a method for sorting.

[0028] In this specification, the user terminal may be a mobile phone, a tablet computer, a personal computer, etc. The server may be a high-performance electronic calculator for storing and processing data. In this specification, the server may store multiple resources to provide the user terminal with background services corresponding to the resources.

[0029] Figure 2 A flowchart of a sorting method provided by an embodiment of the present invention. Figure 2 As shown, the sorting method includes:

[0030] Step 201: obtaining user behavior information of a target user on a product, where the user behavior information includes multiple query items associated with multiple products;

[0031] Step 202: determining a recall query item of the target user according to the similarities between the multiple query items and the target user;

[0032] Step 203: sort the weights of the recalled query items, and display the preferred query items in the recalled query items that meet a preset weight threshold.

[0033] According to the sorting method, apparatus, device and storage medium in the embodiment of the present invention, the recall query items of the target user are determined by calculating the user behavior information including multiple query items associated with the target user and the similarity of the target user; and then the weights of the recall query items are sorted according to the weights of the recall query items, so as to obtain the preference query items that meet the preset weight threshold. The embodiment of the present invention can first filter based on the query items of the recommended products to obtain the recall query items, and then filter the recall query items according to the weights of the recall query items, and recommend the products corresponding to the preference query items that meet the preset weight threshold in the recall query items to the user, so as to achieve accurate recommendation of the products and improve the user experience.

[0034] In one embodiment of the present invention, determining the recall query item of the target user according to the similarity between the multiple query items and the target user includes:

[0035] Taking the node of the target user as the source node of the meta-path, all candidate paths corresponding to the meta-path are obtained from the heterogeneous network, and then the candidate query items corresponding to the end nodes of each candidate path are obtained;

[0036] Determine the similarity between the target user and each candidate query item;

[0037] The candidate query items whose similarity is greater than the corresponding similarity threshold are used as the recalled query items.

[0038] Among them, the similarity threshold can be set based on the recall accuracy, which will not be described here.

[0039] Determining the recall query item of the target user according to the similarity between multiple query items and the target user refers to generating hundreds of query candidates from a large query pool in a heterogeneous network based on meta-paths. The query item refers to the query item through which the corresponding product is displayed on the user terminal, which can be one word or multiple words; in the case of multiple words, each word represents a different query condition, for example, query conditions such as season, size, adult, type information of behavior, time information, etc.

[0040] Correspondingly, in this example, hundreds of query candidates are generated from a large query pool, including:

[0041] Step 1: Convert user behavior information into a heterogeneous network. The heterogeneous network includes path data composed of vertex data and edge data. The vertex data includes users, products associated with user operation behaviors, and query items associated with products. The edge data includes the association relationship between vertex data.

[0042] Step 2: Set the meta-path of vertex data and edge data, and the meta-path starts with the user as the vertex data and ends with the query item as the vertex data;

[0043] Step 3: Calculate the relationship weight between the target user and the query item based on the meta-path.

[0044] The association relationship may include the relationship between the user, the product, and the query item, for example, after the user purchases or searches for a product (user purchase or search behavior data), the purchase association relationship or search association relationship between the user and the product.

[0045] In addition, it should be noted that the association relationship between the vertex data in the embodiment of the present invention is not limited to the above-mentioned purchase association relationship or search association relationship. In practical applications, other association relationships, similar association relationships between commodities, etc. may also be included according to the actual association situation between the vertex data.

[0046] exist Figure 3 In the equation, U2I2Q (i.e. Figure 3 The first element path in Figure 3 The second path in the Figure 3 The third meta-path in the example is used to illustrate. Among them, U is "user", I is "item", S is "scenario", C is "category", and Q is "query". The weight of the meta-path represents the relationship weight Metescore(q) between U and Q as follows:

[0047]

[0048] Among them, U2I2Q: can be quantified into P(q / i), P(q / i) represents the probability of generating a query item (q) given a product i, where P(q / i) is as follows:

[0049]

[0050] Among them, count(q / i) is the record data of i retrieved by searching q, and count(i) is the number of all retrieved records.

[0051] U2I2S2Q: It can be quantified into P(s / i)*P(q / s), where P(s / i) represents the probability of activating scenario S given product i, and P(q / s) represents the probability of generating a query item (q) given scenario (s). * represents multiplication; where S can be the title and product category.

[0052] U2I2C2Q: It can be quantified into P(c / i)*P(q / c), where P(s / i) represents the probability of given product i and c, P(q / c) represents (c) and the probability of generating the query item (q), * represents multiplication, and c can be the category of the product.

[0053] It should be noted that users, items, and queries are three types of vertex data, and there are different interactive relationships between these three types of vertex data. For example, a user directly clicks on an item, a user enters a search by querying a query, and performs item operations in the search.

[0054] In one embodiment of the present invention, the user behavior information further includes feature information of multiple commodities, type information of first historical behaviors associated with the multiple commodities, and time information associated with the multiple commodities. Before ranking the weights of the recalled query items and displaying the preferred query items that meet the preset weight threshold in the recalled query items, the method further includes:

[0055] The weight of the recall query item is determined according to the feature information of the plurality of commodities, the type information of the first historical behavior associated with the plurality of commodities, the time information associated with the plurality of commodities, and the recall query item.

[0056] In this example, according to the feature information of the plurality of commodities, the type information of the first historical behavior associated with the plurality of commodities, the time information associated with the plurality of commodities, and the recall query item, the weight of the recall query item is determined, including:

[0057] Determine first historical behavior characteristic information according to characteristic information of the plurality of commodities, type information of the first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities;

[0058] Calculate the similarity between the recalled query item and the first historical behavior feature information;

[0059] The weights of the recalled query terms are determined based on the similarity.

[0060] exist Figure 4 In the Attention-GRU model, the input sequence x = (x1, x2, xk, ..., xn) is input to obtain the first historical behavior feature information h′ k =f(h k ,A l ,Δt k ); h′ k and Q (i.e. Figure 3 The query candidates in () are used to calculate the weight α1k through the Attention mechanism.

[0061] In this example, determining the first historical behavior characteristic information according to the characteristic information of the plurality of commodities, the type information of the first historical behavior associated with the plurality of commodities, and the time information associated with the plurality of commodities includes:

[0062] Inputting the feature information of the multiple commodities, the type information of the first historical behavior associated with the multiple commodities, and the time information associated with the multiple commodities into the first hidden layer, and obtaining the implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vectors corresponding to the type information of the first historical behavior, and the implicit state vectors corresponding to the time information;

[0063] The implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vector corresponding to the type information of the first historical behavior, and the implicit state vector corresponding to the time information are input into the second hidden layer to obtain the first historical behavior feature information.

[0064] In one example, the characteristic information of multiple commodities refers to the characteristic information for identifying the commodities, and the category of the commodities can be determined; for example, the characteristic information corresponding to a scarf is identified and determined to be a scarf.

[0065] The type information of the first historical behavior associated with the multiple commodities refers to the operation types for the commodities, for example, operation types such as click, browse, collect, purchase, etc.

[0066] The time information associated with multiple products refers to the time difference between the purchase time of the product and the current time. The larger the time difference, the more reference angle of the corresponding query item, that is, the products communicated recently can represent the user's actual demand for the product.

[0067] exist Figure 4 In the Attention-GRU model, the input sequence x = (x1, x2, xk, ..., xn) is passed through the first hidden layer to obtain h = (h1, h2, hk, ..., hn), that is, the hidden state vector. Next, h = (h1, h2, hk, ..., hn) is transformed through the second hidden layer to obtain the first historical behavior feature information h′ k :

[0068] h′ k =f(h k ,A l ,Δt k )

[0069] Where f is the conversion function, h k is the hidden state vector, A l is the type information of the first historical behavior, Δt k For time information.

[0070] In addition, the embodiment of the present invention introduces the Attention mechanism to further determine which factors of each query word (i.e., query condition) in the query item ultimately affect the recommendation accuracy, and then determine the weight of each query word in the query item.

[0071] The Attention mechanism is a concept often used in natural language processing. When performing natural language processing and needing to understand the meaning of a word or a sentence, contextual information is critical and can help understand the exact meaning of a word or a sentence. However, contexts at different positions have different effects on the current word or sentence to be processed, and the "attention" to be invested is also different. In addition, the position of the context that has the greatest impact on the current word or sentence is not fixed, because it may appear before or after the current word or sentence, and the distance is also uncertain. Therefore, the Attention mechanism is needed to solve such problems.

[0072] In one or more embodiments of the present invention, the idea of ​​the Attention mechanism is used to assign a high weight to the query terms in each of the multiple recalled query items that have an important impact on the recommendation results, thereby improving the recommendation accuracy.

[0073] In one embodiment of the present invention, the neural network includes a GRU neural network, a recurrent neural network RNN ​​neural network, or a long short-term memory network LSTM neural network.

[0074] In an example of the present invention, the user behavior information further includes: user basic feature information and environment feature information, and the weights of the recalled query items are sorted, including:

[0075] Determine the order of weights of the recalled query items according to the user basic feature information, the environment feature information, the first historical behavior feature information, and the recalled query items;

[0076] The weights of the recalled query items are sorted according to the order corresponding to the weights of the recalled query items.

[0077] Specifically, the order corresponding to each query item is determined through the user basic feature information, the environment feature information, and the first historical behavior feature information; then the weights of the recalled query items are sorted based on the order corresponding to the weights of the query items. For example, the weights of each query word of each query item are summed, or the weights of each query word are summed according to different preset ratios to obtain the order corresponding to each query item.

[0078] In an example of the present invention, the characteristic information of the training sample further includes: environmental characteristics, and the environmental characteristics specifically include: seasons, festivals, and the like.

[0079] exist Figure 5 In the process, the user's basic feature information, environmental feature information, the first historical behavior feature information and the recalled query items are sequentially input into the fully connected layer and the softmax network layer. Finally, the loss function is calculated by summing up the labels (i.e., the order information) for training until the loss function meets the preset conditions. The final sorting model is obtained, which is used to sort each recalled query item.

[0080] In an example of the present invention, after sorting the weights of the recalled query items and displaying the preferred query items that meet the preset weight threshold in the recalled query items, the sorting method further includes:

[0081] receiving a selection operation of a target query item in the preference query items by a target user;

[0082] Recommend products corresponding to the target query items to the target user based on the selection operation.

[0083] Specifically, after the recall query items are finely sorted, the target query items are displayed on the user terminal, and then the user can select the query item he wants to operate. After the user selects the query item, the product corresponding to the query item is displayed on the user terminal.

[0084] According to an embodiment of the present invention, user behavior information of multiple commodities is combined to allow the recommendation system to better understand the needs of target users. On the other hand, only one interaction is required to achieve rapid recommendation of commodities, thereby achieving fast and accurate recommendation of commodities and improving user experience.

[0085] Figure 6 FIG. 2 shows a schematic diagram of the structure of a recommendation system provided according to an embodiment of the present invention. Figure 6 As shown, the recommendation system includes:

[0086] The question generation module 601 is used to convert the question generation task into a query recommendation task. For example, to recommend four product queries "hat", "scarf", "glove" and "socks" to the user, the corresponding question is "Do you want to buy a hat or scarf or gloves or socks?"

[0087] User feedback module 602, used for user click, browse, purchase and other operations to answer whether the user needs the item, for example, if the user clicks "scarf", it is determined that the user prefers "scarf";

[0088] The project promotion module 603 is used to recommend projects based on user clicks and user behaviors.

[0089] In the embodiment of the present invention, when a user browses items in the recommendation system, the recommendation system generates a question to inquire about the user's interests (i.e., question generation module 601). Once the user's feedback is obtained (i.e., user feedback module 602), the system will recommend related items based on the user's operation mode, that is, the retrieved results are ensured to be closely related to the user's feedback (i.e., query), thereby achieving accurate recommendations (i.e., item promotion module 603).

[0090] Figure 7 FIG. 2 shows a schematic diagram of the structure of a training device for a sorting model according to an embodiment of the present invention. Figure 7 As shown, the training device 700 of the sorting model includes:

[0091] An acquisition module 701 is used to acquire user behavior information of a target user on multiple commodities, where the user behavior information includes multiple query items associated with the multiple commodities;

[0092] A determination module 702, configured to determine a recall query item of the target user according to similarities between the multiple query items and the target user;

[0093] The sorting module 703 is used to sort the weights of the recalled query items and display the preferred query items that meet the preset weight threshold in the recalled query items.

[0094] According to the sorting method, apparatus, device and storage medium in the embodiment of the present invention, the recall query items of the target user are determined by calculating the user behavior information including multiple query items associated with the target user and the similarity of the target user; and then the weights of the recall query items are sorted according to the weights of the recall query items, so as to obtain the preference query items that meet the preset weight threshold. The embodiment of the present invention can first filter based on the query items of the recommended products to obtain the recall query items, and then filter the recall query items according to the weights of the recall query items, and recommend the products corresponding to the preference query items that meet the preset weight threshold in the recall query items to the user, so as to achieve accurate recommendation of the products and improve the user experience.

[0095] Optionally, the determination module 702 is further configured to:

[0096] Taking the node of the target user as the source node of the meta-path, all candidate paths corresponding to the meta-path are obtained from the heterogeneous network, and then the candidate query items corresponding to the end nodes of each candidate path are obtained;

[0097] Determine the similarity between the target user and each candidate query item;

[0098] The candidate query items whose similarity is greater than the corresponding similarity threshold are used as the recalled query items.

[0099] Optionally, the user behavior information further includes feature information of multiple commodities, type information of the first historical behavior associated with the multiple commodities, and time information associated with the multiple commodities. The determination module 702 is further configured to:

[0100] The weight of the recall query item is determined according to the feature information of the plurality of commodities, the type information of the first historical behavior associated with the plurality of commodities, the time information associated with the plurality of commodities, and the recall query item.

[0101] Optionally, the determination module 702 is further configured to:

[0102] Determine first historical behavior characteristic information according to characteristic information of the plurality of commodities, type information of the first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities;

[0103] Calculate the similarity between the recalled query item and the first historical behavior feature information;

[0104] The weights of the recalled query terms are determined based on the similarity.

[0105] Optionally, the determination module 702 is further configured to:

[0106] Inputting feature information of the plurality of commodities, type information of the first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities into a first hidden layer, and obtaining an implicit state vector corresponding to the feature information of the plurality of commodities, an implicit state vector corresponding to the type information of the first historical behavior, and an implicit state vector corresponding to the time information;

[0107] The implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vectors corresponding to the type information of the first historical behavior, and the implicit state vectors corresponding to the time information are input into the second hidden layer to obtain the first historical behavior feature information.

[0108] Optionally, the user behavior information further includes: user basic feature information and environment feature information. The sorting module 703 is further used to:

[0109] Determining the order of the weights of the recall query items according to the user basic feature information, the environment feature information, the first historical behavior feature information, and the weights of the recall query items;

[0110] The weights of the recalled query items are sorted according to the order corresponding to the weights of the recalled query items.

[0111] Optionally, the sorting device further includes:

[0112] A receiving module, used for receiving a selection operation of a target user on a target query item in a preference query item;

[0113] The recommendation module is used to recommend products corresponding to the target query items to the target user according to the selection operation.

[0114] Figure 8 FIG. 8 is a schematic diagram showing a hardware structure 800 of a computing device according to an embodiment of the invention.

[0115] like Figure 8 As shown, the computing device 800 includes an input device 801, an input interface 802, a central processing unit 803, a memory 804, an output interface 805, and an output device 806. The input interface 802, the central processing unit 803, the memory 804, and the output interface 805 are connected to each other via a bus 86, and the input device 801 and the output device 806 are connected to the bus 86 via the input interface 802 and the output interface 805, respectively, and then connected to other components of the computing device 800.

[0116] Specifically, the input device 801 receives input information from the outside and transmits the input information to the central processing unit 803 through the input interface 802; the central processing unit 803 processes the input information based on the computer executable instructions stored in the memory 804 to generate output information, temporarily or permanently stores the output information in the memory 804, and then transmits the output information to the output device 806 through the output interface 805; the output device 806 outputs the output information to the outside of the computing device 800 for user use.

[0117] That is to say, Figure 8 The computing device shown may also be implemented as including: a memory storing computer executable instructions; and a processor, which can implement the combination of Figure 2 , Figure 7 The method and apparatus described.

[0118] An embodiment of the present invention further provides a computer storage medium, on which computer program instructions are stored; when the computer program instructions are executed by a processor, the information processing method provided by the embodiment of the present invention is implemented.

[0119] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. Programs or code segments can be stored in machine-readable media, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable media" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0120] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0121] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be covered within the protection scope of the present invention.

Claims

1. A sorting method, the method comprising: Acquire user behavior information of a target user on a plurality of commodities, the user behavior information including a plurality of query items associated with the plurality of commodities, feature information of the plurality of commodities, type information of a first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities; Determining a recall query item of the target user according to similarities between the multiple query items and the target user; determining a weight of the recall query item according to feature information of the plurality of commodities, type information of the first historical behaviors associated with the plurality of commodities, time information associated with the plurality of commodities, and the recall query item; The weights of the recalled query items are sorted, and the preferred query items in the recalled query items that meet a preset weight threshold are displayed.

2. The method according to claim 1, wherein: The determining, according to the similarities between the multiple query items and the target user, the recalled query items of the target user comprises: Taking the node of the target user as the source node of the meta-path, obtaining all candidate paths corresponding to the meta-path from the heterogeneous network, and then obtaining candidate query items corresponding to the end nodes of each candidate path; Determining the similarity between the target user and each candidate query item; The candidate query items whose similarity is greater than the corresponding similarity threshold are used as the recalled query items.

3. The method according to claim 1, wherein: The determining the weight of the recall query item according to the feature information of the plurality of commodities, the type information of the first historical behavior associated with the plurality of commodities, the time information associated with the plurality of commodities, and the recall query item includes: Determine first historical behavior characteristic information according to the characteristic information of the plurality of commodities, type information of the first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities; Calculating the similarity between the recall query item and the first historical behavior feature information; A weight of the recalled query term is determined based on the similarity.

4. The method according to claim 3, wherein: The determining the first historical behavior characteristic information according to the characteristic information of the plurality of commodities, the type information of the first historical behavior associated with the plurality of commodities, and the time information associated with the plurality of commodities includes: Inputting the feature information of the multiple commodities, the type information of the first historical behavior associated with the multiple commodities, and the time information associated with the multiple commodities into the first hidden layer, and obtaining the implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vectors corresponding to the type information of the first historical behavior, and the implicit state vectors corresponding to the time information; The implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vector corresponding to the type information of the first historical behavior, and the implicit state vector corresponding to the time information are input into the second hidden layer to obtain the first historical behavior feature information.

5. The method according to any one of claims 1 to 4, wherein: The user behavior information also includes: basic user feature information and environmental feature information. The weights of the recalled query items are sorted, including: Determining the order of the weights of the recall query items according to the user basic feature information, the environment feature information, and the first historical behavior feature information; The weights of the recall query items are sorted according to the order corresponding to the weights of the recall query items.

6. The method according to claim 1, wherein: After ranking the weights of the recalled query items and displaying the preferred query items that meet a preset weight threshold among the recalled query items, the method further includes: receiving a selection operation of a target user on a target query item in the preference query items; The commodity corresponding to the target query item is recommended to the target user according to the selection operation.

7. A sorting device, comprising: an acquisition module, configured to acquire user behavior information of a target user on a plurality of commodities, the user behavior information comprising a plurality of query items associated with the plurality of commodities, feature information of the plurality of commodities, type information of a first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities; a determination module, configured to determine a recall query item of the target user according to the similarities between the multiple query items and the target user, and determine a weight of the recall query item according to the feature information of the multiple commodities, the type information of the first historical behaviors associated with the multiple commodities, the time information associated with the multiple commodities, and the recall query item; The sorting module is used to sort the weights of the recalled query items and display the preferred query items in the recalled query items that meet a preset weight threshold.

8. The device according to claim 7, wherein: The determining module is further used for: Taking the node of the target user as the source node of the meta-path, obtaining all candidate paths corresponding to the meta-path from the heterogeneous network, and then obtaining candidate query items corresponding to the end nodes of each candidate path; Determining the similarity between the target user and each candidate query item; The candidate query items whose similarity is greater than the corresponding similarity threshold are used as the recalled query items.

9. The device according to claim 7, wherein: The determining module is further used for: Determine first historical behavior characteristic information according to the characteristic information of the plurality of commodities, type information of the first historical behavior associated with the plurality of commodities, and time information associated with the plurality of commodities; Calculating the similarity between the recall query item and the first historical behavior feature information; A weight of the recalled query term is determined based on the similarity.

10. The device according to claim 9, wherein: The determining module is further used for: Inputting the feature information of the multiple commodities, the type information of the first historical behavior associated with the multiple commodities, and the time information associated with the multiple commodities into the first hidden layer, and obtaining the implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vectors corresponding to the type information of the first historical behavior, and the implicit state vectors corresponding to the time information; The implicit state vectors corresponding to the feature information of the multiple commodities, the implicit state vector corresponding to the type information of the first historical behavior, and the implicit state vector corresponding to the time information are input into the second hidden layer to obtain the first historical behavior feature information.

11. The device according to any one of claims 7 to 10, wherein: User behavior information also includes: user basic feature information and environmental feature information. The sorting module is also used to: Determining the order of the weights of the recall query items according to the user basic feature information, the environment feature information, the first historical behavior feature information, and the weights of the recall query items; The weights of the recall query items are sorted according to the order corresponding to the weights of the recall query items.

12. The device according to claim 7, wherein: The device also includes: A receiving module, used for receiving a selection operation of a target user on a target query item in the preference query items; A recommendation module is used to recommend the commodity corresponding to the target query item to the target user according to the selection operation.

13. A computing device, wherein: The computing device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method according to any one of claims 1 to 6 is implemented.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes instructions, and when the instructions are executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 6.

Images