Method and apparatus for ranking multiple recall results

Abstract

This specification provides a method and apparatus for sorting multi-path recall results. The sorting method for multi-path recall results includes: receiving a recommendation request and obtaining target user characteristics based on the recommendation request; recalling related item information of at least one related dimension based on the target user characteristics, and determining the recall association weight and recall screening weight of the related item information; obtaining target related item information from the related item information according to the recall screening weight; and sorting the target related item information based on the target related item information and the recall association weight to obtain a related item sequence.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method for ranking multi-channel recall results. One or more embodiments of this specification also relate to a multi-channel recall result ranking apparatus, a recommendation system, a computing device, a computer-readable storage medium, and a computer program. Background Technology

[0002] In order to maintain the platform's attractiveness and improve the user experience, many platforms often provide personalized recommendations for different users, enabling them to quickly find content that matches their interests. To achieve personalized push notifications, recommendation systems are often used. Recommendation systems recall content that users may be interested in based on their historical behavior and recommend it to them.

[0003] However, the amount of data recalled by current recommendation systems is unstable, and the sorting of the recalled data is not accurate enough, resulting in serious homogenization and poor recommendation performance. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a method for ranking multi-path recall results. One or more embodiments of this specification also relate to a device for ranking multi-path recall results, a recommendation system, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a method for sorting multi-path recall results is provided, comprising:

[0006] Receive recommendation requests and obtain target user characteristics based on the recommendation requests;

[0007] Based on the target user characteristics, recall related item information of at least one related dimension, and determine the recall association weight and recall screening weight of the related item information;

[0008] Based on the recall screening weights, the target associated project information is obtained from the associated project information;

[0009] The target associated project information is sorted based on the target associated project information and the recall associated weight to obtain a sequence of associated projects.

[0010] According to a second aspect of the embodiments of this specification, a sorting apparatus for multi-channel recall results is provided, comprising:

[0011] The receiving module is configured to receive recommendation requests and obtain target user characteristics based on the recommendation requests;

[0012] The determination module is configured to recall associated item information of at least one associated dimension based on the target user characteristics, and to determine the recall association weight and recall screening weight of the associated item information.

[0013] The acquisition module is configured to acquire target associated project information from the associated project information according to the recall screening weight;

[0014] The sorting module is configured to sort the target associated project information based on the target associated project information and the recall associated weight to obtain a sequence of associated projects.

[0015] According to a third aspect of the embodiments of this specification, a recommendation system is provided, the recommendation system including a recall module, a coarse ranking module, a recall merging module, and a fine ranking module, wherein:

[0016] The recall module receives a recommendation request and obtains target user characteristics based on the recommendation request; it then recalls related item information for at least one related dimension based on the target user characteristics.

[0017] The coarse sorting module sorts the associated project information based on a preset sorting strategy;

[0018] The recall merging module determines the recall association weight and recall screening weight of the associated project information based on the associated project information of the at least one association dimension, and obtains the target associated project information from the associated project information according to the recall screening weight; and sorts the target associated project information based on the target associated project information and the recall association weight to obtain the associated project sequence.

[0019] The fine ranking module acquires a set of feature information and generates a recommendation sequence based on the associated project sequence and the set of feature information.

[0020] According to a fourth aspect of the embodiments of this specification, a computing device is provided, including a memory, a processor, and computer instructions stored in the memory and executable on the processor, wherein the processor, when executing the computer instructions, implements the steps of the sorting method for multiplexed recall results.

[0021] According to a fifth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer instructions, which, when executed by a processor, implement the steps of the sorting method for the multiplexed recall results.

[0022] According to a sixth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the sorting method for the results of the multiple-way recall described above.

[0023] The multi-path recall result ranking method provided in this specification involves receiving a recommendation request and obtaining target user characteristics based on the recommendation request; recalling related item information of at least one related dimension based on the target user characteristics, and determining the recall association weight and recall screening weight of the related item information; obtaining target related item information from the related item information according to the recall screening weight; and ranking the target related item information based on the target related item information and the recall association weight to obtain a related item sequence.

[0024] One embodiment of this specification implements the determination of recall association weights and recall screening weights through associated item information, thereby facilitating the identification of target associated item information from associated item information and increasing the correlation between the recalled associated item information and the characteristics of the target user; based on the target recall association weights, the determined target associated item information is sorted to obtain an associated item sequence, thereby increasing the sorting accuracy of the associated item sequence. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method for sorting multi-path recall results according to one embodiment of this specification;

[0026] Figure 2 This is a flowchart illustrating the processing procedure of a method for sorting multi-way recall results provided in one embodiment of this specification.

[0027] Figure 3 This is a schematic diagram of the structure of a recommendation system provided in one embodiment of this specification;

[0028] Figure 4 This is a schematic diagram of a sorting device for multi-path recall results provided in one embodiment of this specification;

[0029] Figure 5 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0030] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0031] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to any or all possible combinations including one or more of the associated listed items.

[0032] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0033] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0034] Recall: Collecting relevant objects (such as products, general merchandise, text, etc.) upon user request.

[0035] Multi-path recall merging: The recall results from different recall methods are merged and truncated before being sent to downstream modules.

[0036] Sort-aggregate: A combinatorial mathematics problem that involves merging n homogeneous sorts and optimizing them within a certain metric space.

[0037] Multi-path recall merging is a crucial issue encountered in large-scale search and recommendation scenarios. Typically, multi-path recall merging involves two parts: 1. Truncation of each recall source to limit the number of recalls, which is done manually; 2. Feeding the truncated results into a coarse-grained ranking model, usually a shallow, high-speed model.

[0038] Currently, most recommendation systems reuse the above process, but the above technology has obvious drawbacks: 1. The recall quantity is inflexible and requires manual maintenance; 2. Shallow models mostly depict linear relationships, resulting in serious homogenization of coarse-ranked outputs, often requiring additional diversity plugins; 3. The suboptimal nature of two-stage optimization is obvious, producing serious edge effects.

[0039] To address the aforementioned issues, this specification provides a method for ranking multi-path recall results. This method sorts and aggregates the recall data after coarse ranking or the recall data before coarse ranking, thereby improving the recall merging effect and ultimately enhancing the recommendation accuracy of the recommendation system.

[0040] This specification provides a method for sorting multiple recall results. It also relates to a sorting device for multiple recall results, a computing device, a computer-readable storage medium, and a computer program, which will be described in detail in the following embodiments.

[0041] Figure 1 A flowchart of a method for sorting multiple recall results according to an embodiment of this specification is shown, including steps 102 to 108.

[0042] Step 102: Receive a recommendation request and obtain the target user characteristics based on the recommendation request.

[0043] Specifically, the recommendation system receives recommendation requests, which are requests to a target platform to recommend data content associated with user characteristics. The target platform refers to a platform that has recommendation needs, such as a music platform, video platform, shopping platform, etc. In practical applications, the target platform can issue recommendation requests to target users, or the user can trigger the generation of recommendation requests. This application does not make specific limitations.

[0044] For example, a recommendation request is a request issued by a music platform to recommend relevant music to user A; another example is a recommendation request issued by a shopping platform to recommend relevant products to user B, and so on.

[0045] After receiving a recommendation request, the recommendation platform can obtain the target user's characteristics based on the recommendation request, which can be used for subsequent recommendations based on the target user's characteristics. Here, user characteristics refer to historical or real-time data related to the user. For example, user A's user characteristic information is that he bought a cup of brand G in store D. Another example is that user B listened to a song with the name H in music platform S.

[0046] In practical applications, target user characteristics can be obtained in the following ways. Specifically, the method for obtaining target user characteristics based on the recommendation request may include:

[0047] Parse the recommendation request to obtain the target user identifier;

[0048] Based on the target user identifier, obtain the target user features from the feature database.

[0049] Specifically, parsing the recommendation request can retrieve the target user identifier contained within it. The target user identifier refers to fields that identify the target user, such as user ID, username, and IP address. After obtaining the target user identifier, the corresponding target user features can be retrieved from a feature database. This feature database stores user features; with the user's consent to the retrieval of their features, the user's feature information can be stored, thus obtaining the feature database. In practical applications, user feature information can also be obtained through other methods; this application does not impose specific limitations on these methods.

[0050] For example, the recommendation system receives a music recommendation request for user A; it parses the music recommendation request to obtain user A's user ID "123"; and it queries the feature data to find the user features corresponding to the user ID. The user features include user A's viewing history on video platforms, listening history on music platforms, etc.

[0051] By receiving recommendation requests, target user feature information is obtained, thereby determining the recommended objects and the basis for the recommendations, which facilitates subsequent retrieval of relevant data based on target user feature information.

[0052] Step 104: Based on the target user characteristics, recall related item information of at least one related dimension, and determine the recall association weight and recall screening weight of the related item information.

[0053] After identifying the characteristics of the target user, information related to the target user characteristics can be recalled. For example, based on user A's user characteristic "listening to singer A's song 1 on a music platform", information related to the user characteristics such as "song 2 by singer A" and "song 1 sung by singer B" can be recalled.

[0054] Among them, the association dimension refers to the semantic perspective determined based on user feature information, such as the product dimension, product brand dimension, etc.; the associated item information refers to the item information that is associated with the target user's features; the recall association weight refers to the association weight corresponding to each association dimension. For example, the recommendation system recalls associated item information from 3 dimensions, and the recall association weights for each dimension are 0.4, 0.3, and 0.3, respectively; the recall screening weight refers to the weight when screening associated item information under each association dimension. For example, the recall screening weight of association dimension F is 0.8, that is, based on the weight of 0.8, 400 associated item information can be determined from 500 associated item information under association dimension F.

[0055] In practical applications, various recall strategies can be used to recall related project information, thereby ensuring the richness of related project information and thus ensuring the accuracy of subsequent recommendations.

[0056] Specifically, a method for recalling associated item information of at least one related dimension based on the target user characteristics may include:

[0057] Parse the target user features to obtain at least one related dimension feature;

[0058] Retrieve associated project information corresponding to each associated dimension based on the features of each associated dimension.

[0059] Among them, the related dimension features refer to the feature information obtained by parsing the target user features. For example, based on preset semantic analysis rules, multiple related dimension features can be extracted from the target user features. Specifically, the feature "purchase a cup of brand b from store A" can be extracted into three related dimension features: store name, brand name, and product name based on semantic analysis rules. The related dimensions are then determined based on these related dimension features, and the corresponding related item information under each related dimension can be obtained. Furthermore, the related dimensions can be set according to actual recommendation needs; this application does not impose specific limitations on this.

[0060] For example, user D's user characteristic is "listened to singer B's song g on music platform M"; based on the preset parsing rules, user D's user characteristic information is parsed to determine three dimensions: "music platform, singer name, and song name"; according to the user characteristics, the associated music platform information under the music platform dimension, the associated singer name information under the singer name dimension, and the associated song name information under the song name dimension are obtained respectively.

[0061] By acquiring information about related items in each dimension, we can enrich the content of the recommendation information, thereby identifying more relevant related items and improving the accuracy of subsequent recommendations.

[0062] In practical applications, the method for determining the recall association weight and recall screening weight of the associated project information may include:

[0063] Calculate the recall association weight corresponding to each association dimension based on the target user characteristics and the associated item information under each association dimension.

[0064] Determine the time attribute information in the associated project information under each associated dimension, and calculate the first recall screening weight corresponding to the associated project information based on each time attribute information;

[0065] Determine the degree of association information in the associated project information under each association dimension, and calculate the second recall screening weight corresponding to the associated project information based on each degree of association information.

[0066] Specifically, user characteristics are compared with the associated item information under each associated dimension to calculate the matching degree information between user characteristics and each associated item information; the recall association weight corresponding to each associated dimension is calculated according to the preset weight calculation rules and each matching degree information. The preset weight rules can be set based on actual needs.

[0067] For example, if it is determined that there are related item information A and related item information B under the association dimension A, and the matching degree information between user feature N and related information A and related information B is 40% and 50% respectively, and the preset weight calculation rule is to calculate the average value of the matching degree information, then the average of the matching degree information of 40% and 50% is 45%, that is, the recall association weight of association dimension A is 45%.

[0068] Determine the time attribute information in the associated project information. The time attribute information refers to the time attribute information contained in the associated project information that corresponds to the associated project information. For example, if a user's characteristic is that they watch live broadcasts three times a week between 9 pm and 10 pm, then the associated project information containing time attribute information can be determined based on the time information in the user's characteristics, such as broadcasting four times a week between 9 pm and 10 pm. Determine the associated project information containing time attribute information under each dimension, and calculate the first recall screening weight based on the number of associated project information containing time attribute information under each associated dimension.

[0069] For example, if the total number of associated item information under association dimension B is determined to be 500, and the total number of associated items containing time attribute information is determined to be 100, then the weight of the first recall quantity corresponding to the associated item information under association dimension B is calculated to be 0.25 based on the total number of associated item information of 500 and the total number of associated items containing time attribute information of 100.

[0070] Determine the degree of association information in the associated project information, where the degree of association information refers to the degree of association information corresponding to the associated project information contained in the associated project information; determine the associated project information containing degree of association information under each dimension, and calculate the second recall screening weight based on the number of associated project information containing associated dimension information under each associated dimension.

[0071] For example, if the total number of associated item information under association dimension B is determined to be 500, and the total number of associated items containing association degree information is determined to be 100, then the second recall quantity weight corresponding to the associated item information under association dimension B is calculated to be 0.25 based on the total number of associated item information of 500 and the total number of associated items containing association degree information of 100.

[0072] By using the recall association weight and recall filtering weight of associated project information, it is easier to filter target associated project information from the associated project information in the future.

[0073] Step 106: Obtain the target related project information from the related project information according to the recall screening weight.

[0074] After determining the recall screening weights, the associated project information under each association dimension is screened based on the recall screening weights to obtain the target associated project information under each association dimension.

[0075] For example, if the associated item information under association dimension A is determined to be {a, c, b, d}, and the recall screening weight corresponding to association dimension A is 0.5, and the total number of associated item information under association dimension A is 4, then the screening quantity is determined to be 2. In this case, 2 associated item information can be selected from the associated item information as the target associated item information.

[0076] Step 108: Sort the target associated project information based on the target associated project information and the recall associated weight to obtain the associated project sequence.

[0077] After determining the target associated item information and the target recall associated weight, the target associated item information and the target recall associated weight can be input into the sorting and aggregation module, and the associated item sequence output by the sorting and aggregation module can be obtained.

[0078] In practical applications, a method for sorting the target associated item information based on the target associated item information and the recall associated weight to obtain a sequence of associated items may include:

[0079] A preset sorting strategy is determined, and the target associated item information under each associated dimension is sorted based on the preset sorting strategy to obtain the target associated item sequence corresponding to each associated dimension.

[0080] Each target-related item sequence is sorted based on the target recall association weight corresponding to each target-related item sequence to obtain the related item sequence.

[0081] Among them, the preset ranking strategy refers to the strategy for ranking target related items. In practical applications, the preset ranking strategy can be a coarse ranking model, which is used for simple screening and recall of target related item information; the target related item association sequence refers to the sequence obtained by ranking the target related item information; each target related item sequence obtained by ranking and the corresponding target recall weight are input into the ranking aggregation model to obtain the related item sequence output by the ranking aggregation model.

[0082] In practical applications, the sorting aggregation model contains preset sorting aggregation rules, and can sort the target associated item sequence based on the preset sorting aggregation rules.

[0083] Specifically, sorting each target associated item sequence based on the target recall association weight corresponding to each target associated item sequence, the method for obtaining the associated item sequence may include:

[0084] Determine the preset sorting aggregation rules;

[0085] Based on the preset sorting rules, each target associated item sequence, and each target associated item sequence, obtain the associated item sequence.

[0086] Among them, the preset sorting aggregation rule refers to the rule that can perform aggregation sorting on the target associated item sequence. For example, the following formula 1 can be used as the preset sorting aggregation rule:

[0087]

[0088] Among them, pi represents the i-th target associated item sequence, n is the total number of target associated item sequences. If a target associated item sequence has m elements, C represents the set of target associated item sequences of all permutations of these m elements, wi represents the weight of the i-th target associated item sequence, and dτ(p, q) represents the Kendall Tau distance.

[0089] In addition, the preset sorting aggregation rule can also be the Borda count method, the Lehmer encoding and decoding method, the Copeland method, etc.; among them, the calculation method of the Borda count method is to calculate the average position of each element in the target associated item sequence, and use this as the position of the element in the associated item sequence; the calculation method of the Copeland method is: when i < j means that more than half of the associated item sequences rank element i in front of element j, i is the predecessor of j, j is the successor of i, and then define the Copeland score of element i as |{j:i<j}|, that is, the number of successors of element i. Finally, sort the elements in descending order according to the Copeland score; the calculation method of the Lehmer encoding and decoding method is: for each element, calculate its Lehmer encoding in all input permutations, obtain the mode of the encoding result as the encoding of the element, and perform Lehmer decoding on the encoded result to obtain the associated item sequence.

[0090] Preferably, in order to ensure the accuracy of the output result of the sorting aggregation model, the parameters of the sorting aggregation model can be optimized based on user characteristics, so as to obtain a sorting aggregation model with higher output result accuracy.

[0091] This specification provides a method for ranking multi-path recall results, which involves receiving a recommendation request and obtaining target user characteristics based on the recommendation request; recalling related item information of at least one related dimension based on the target user characteristics, and determining the recall association weight and recall screening weight of the related item information; obtaining target related item information from the related item information according to the recall screening weight; and ranking the target related item information based on the target related item information and the recall association weight to obtain a related item sequence. One embodiment of this specification implements the determination of recall association weight and recall screening weight through related item information, thereby facilitating the identification of target related item information from related item information and increasing the correlation between the recalled related item information and the target user characteristics; ranking the determined target related item information based on the target recall association weight to obtain a related item sequence increases the ranking accuracy of the related item sequence.

[0092] The following is in conjunction with the appendix Figure 2 Taking the sorting method for multi-channel recall results provided in this specification in the application of product recommendation as an example, the sorting method for multi-channel recall results will be further explained. Among them, Figure 2 The flowchart of a sorting method for multi-way recall results provided in one embodiment of this specification is shown, with specific steps including steps 202 to 212.

[0093] Step 202: Receive the recommendation request and obtain the user's purchase characteristic information based on the recommendation request.

[0094] Specifically, purchase feature information includes product features, product brand features, and product store features, etc.

[0095] Step 204: Determine related dimensions such as product dimension, product brand dimension, and product store dimension based on purchase feature information, and obtain related item information under each related dimension.

[0096] Step 206: Input the purchase feature information into the parameter generation model.

[0097] Step 208: Generate model output parameters based on the parameters, and obtain the target associated project information from the associated project information.

[0098] Specifically, the model output parameters include recall association weights and recall filtering weights, and target associated project information is filtered from the associated project information based on the recall filtering weights.

[0099] Step 210: Input the target associated project information and model output parameters into the sorting aggregation model to obtain the associated project sequence.

[0100] Specifically, the recall association weights and target association item information from the model output parameters are input into the ranking aggregation model to obtain the output result of the ranking aggregation model, namely the association item sequence.

[0101] Step 212: Input the sequence of related items into the downstream module for further sorting to obtain recommendation results.

[0102] Specifically, downstream modules can be fine-ranking modules or other modules that can further process the related item sequences to obtain recommendation results.

[0103] Step 214: Feedback the recommendation results to the user.

[0104] Specifically, the recommendation results will be fed back to the user, who can view the product information recommended by the recommendation system.

[0105] The multi-path recall result ranking method provided in this specification involves receiving a recommendation request and obtaining target user characteristics based on the recommendation request; recalling related item information of at least one related dimension based on the target user characteristics, and determining the recall association weight and recall screening weight of the related item information; obtaining target related item information from the related item information according to the recall screening weight; and ranking the target related item information based on the target related item information and the recall association weight to obtain a related item sequence.

[0106] One embodiment of this specification implements the determination of recall association weights and recall screening weights through associated item information, thereby facilitating the identification of target associated item information from associated item information and increasing the correlation between the recalled associated item information and the characteristics of the target user; based on the target recall association weights, the determined target associated item information is sorted to obtain an associated item sequence, thereby increasing the sorting accuracy of the associated item sequence.

[0107] Figure 3 This specification illustrates a recommendation system according to an embodiment, the recommendation system including a recall module, a coarse ranking module, a recall merging module, and a fine ranking module, wherein:

[0108] The recall module 302 receives a recommendation request and obtains target user characteristics based on the recommendation request; and recalls related item information of at least one related dimension based on the target user characteristics.

[0109] The coarse sorting module 304 sorts the associated project information based on a preset sorting strategy;

[0110] The recall merging module 306 determines the recall association weight and recall screening weight of the associated project information based on the associated project information of the at least one association dimension, and obtains the target associated project information from the associated project information according to the recall screening weight; and sorts the target associated project information based on the target associated project information and the recall association weight to obtain the associated project sequence.

[0111] The fine ranking module 308 acquires a set of feature information and generates a recommendation sequence based on the associated project sequence and the set of feature information.

[0112] The feature information set refers to the set of feature information; in the fine ranking, the related item sequence is further sorted according to the feature information to obtain the recommendation sequence used for recommendation.

[0113] The recommendation system provided in this manual adds a recall merging module between the coarse-ranking module and the fine-ranking module. This overcomes the problem of homogenized output results caused by only performing coarse-ranking, increases the relevance of the recall information, and thus increases the recommendation accuracy of the recommendation system.

[0114] Corresponding to the above method embodiments, this specification also provides embodiments of a sorting device for multi-channel recall results. Figure 4 A schematic diagram of a sorting device for multi-path recall results provided in one embodiment of this specification is shown. Figure 4 As shown, the device includes:

[0115] The receiving module 402 is configured to receive a recommendation request and obtain target user characteristics based on the recommendation request;

[0116] The determination module 404 is configured to recall associated item information of at least one associated dimension based on the target user characteristics, and to determine the recall association weight and recall screening weight of the associated item information.

[0117] The acquisition module 406 is configured to acquire target related project information from the related project information according to the recall screening weight;

[0118] The sorting module 408 is configured to sort the target associated project information based on the target associated project information and the recall associated weight to obtain a sequence of associated projects.

[0119] Optionally, the receiving module 402 is further configured to:

[0120] Parse the recommendation request to obtain the target user identifier;

[0121] Based on the target user identifier, obtain the target user features from the feature database.

[0122] Optionally, the determining module 404 is further configured to:

[0123] Parse the target user features to obtain at least one related dimension feature;

[0124] Retrieve associated project information corresponding to each associated dimension based on the features of each associated dimension.

[0125] Optionally, the determining module 404 is further configured to:

[0126] Calculate the recall association weight corresponding to each association dimension based on the target user characteristics and the associated item information under each association dimension.

[0127] Determine the time attribute information in the associated project information under each associated dimension, and calculate the first recall screening weight corresponding to the associated project information based on each time attribute information;

[0128] Determine the degree of association information in the associated project information under each association dimension, and calculate the second recall screening weight corresponding to the associated project information based on each degree of association information.

[0129] Optionally, the sorting module 408 is further configured to:

[0130] A preset sorting strategy is determined, and the target associated item information under each associated dimension is sorted based on the preset sorting strategy to obtain the target associated item sequence corresponding to each associated dimension.

[0131] Each target-related item sequence is sorted based on the target recall association weight corresponding to each target-related item sequence to obtain the related item sequence.

[0132] Optionally, the sorting module 408 is further configured to:

[0133] Define the preset sorting and aggregation rules;

[0134] The associated item sequence is obtained based on the preset sorting rules, each target associated item sequence, and each target associated item sequence.

[0135] This specification provides a multi-path recall result sorting device that receives a recommendation request and obtains target user characteristics based on the recommendation request; recalls related item information of at least one related dimension based on the target user characteristics, and determines the recall association weight and recall screening weight of the related item information; obtains target related item information from the related item information according to the recall screening weight; and sorts the target related item information based on the target related item information and the recall association weight to obtain a related item sequence. One embodiment of this specification implements the determination of recall association weight and recall screening weight through related item information, thereby facilitating the determination of target related item information from related item information and increasing the correlation between the recalled related item information and the target user characteristics; and sorts the determined target related item information based on the target recall association weight to obtain a related item sequence, increasing the sorting accuracy of the related item sequence.

[0136] The above is a schematic scheme of a sorting device for multi-channel recall results according to this embodiment. It should be noted that the technical solution of this sorting device for multi-channel recall results belongs to the same concept as the technical solution of the sorting method for multi-channel recall results described above. For details not described in detail in the technical solution of the sorting device for multi-channel recall results, please refer to the description of the technical solution of the sorting method for multi-channel recall results described above.

[0137] Figure 5 A structural block diagram of a computing device 500 according to an embodiment of this specification is shown. The components of the computing device 500 include, but are not limited to, a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and a database 550 is used to store data.

[0138] The computing device 500 also includes an access device 540, which enables the computing device 500 to communicate via one or more networks 560. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0139] In one embodiment of this specification, the above-described components of the computing device 500 and Figure 5 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 5The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0140] The computing device 500 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device 500 can also be a mobile or stationary server.

[0141] The processor 520 executes the computer instructions to implement the sorting method for the multi-way recall results.

[0142] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above-described method for sorting multiple recall results belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-described method for sorting multiple recall results.

[0143] An embodiment of this specification also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the sorting method for multiplexed recall results as described above.

[0144] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the above-described method for sorting multiple recall results. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described method for sorting multiple recall results.

[0145] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the sorting method for the results of the multiple-way recall described above.

[0146] The above is an illustrative example of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the technical solution of the above-described method for ranking multi-way recall results. Details not described in detail in the computer program's technical solution can be found in the description of the technical solution of the above-described method for ranking multi-way recall results.

[0147] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0148] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0149] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0150] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0151] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for ranking multi-path recall results, comprising: Receive recommendation requests and obtain target user characteristics based on the recommendation requests; Based on the target user characteristics, recall at least one related item information of related dimensions, and determine the recall association weight and recall filtering weight of the related item information. The recall filtering weight is determined based on the time attribute information and the degree of association information in the related item information. The recall association weight refers to the association weight corresponding to each related dimension, and the recall filtering weight refers to the weight when filtering related item information under each related dimension. Based on the recall screening weights, the target associated project information is obtained from the associated project information; The target associated project information is sorted based on the target associated project information and the recall associated weight to obtain an associated project sequence. The sorting includes: according to a preset sorting and aggregation rule, the target associated project sequence composed of the target associated project information selected under each associated dimension is sorted and aggregated in combination with the target recall associated weight corresponding to each target associated project sequence to obtain an associated project sequence.

2. The method as described in claim 1, wherein obtaining target user characteristics based on the recommendation request includes: Parse the recommendation request to obtain the target user identifier; Based on the target user identifier, obtain the target user features from the feature database.

3. The method as described in claim 1, wherein recalling associated item information based on at least one associated dimension according to the target user characteristics includes: Parse the target user features to obtain at least one related dimension feature; Retrieve associated project information corresponding to each associated dimension based on the features of each associated dimension.

4. The method as described in claim 1, wherein sorting the target associated project information based on the target associated project information and the recall associated weight to obtain an associated project sequence includes: A preset sorting strategy is determined, and the target associated item information under each associated dimension is sorted based on the preset sorting strategy to obtain the target associated item sequence corresponding to each associated dimension. Each target-related item sequence is sorted based on the target recall association weight corresponding to each target-related item sequence to obtain the related item sequence.

5. The method as described in claim 4, wherein sorting each target associated item sequence based on the target recall association weight corresponding to each target associated item sequence to obtain the associated item sequence includes: Define the preset sorting and aggregation rules; The associated item sequence is obtained based on the preset sorting and aggregation rules, each target associated item sequence, and each target associated item sequence.

6. A recommendation system, the recommendation system comprising a recall module, a coarse ranking module, a recall merging module, and a fine ranking module, wherein: The recall module receives recommendation requests and obtains target user characteristics based on the recommendation requests; Based on the target user characteristics, recall related item information in at least one related dimension; The coarse sorting module sorts the associated project information based on a preset sorting strategy; The recall merging module determines the recall association weight and recall screening weight of the associated project information based on the associated project information of the at least one association dimension, and obtains the target associated project information from the associated project information according to the recall screening weight. The target associated project information is sorted based on the target associated project information and the recall associated weight to obtain an associated project sequence. The recall screening weight is determined based on the time attribute information and the degree of association information in the associated project information. The recall associated weight refers to the association weight corresponding to each association dimension. The recall screening weight refers to the weight when screening associated project information under each association dimension. The sorting includes: according to a preset sorting aggregation rule, the target associated project sequence composed of the target associated project information screened under each association dimension is sorted and aggregated in combination with the target recall associated weight corresponding to each target associated project sequence to obtain an associated project sequence. The fine ranking module acquires a set of feature information and generates a recommendation sequence based on the associated project sequence and the set of feature information.

7. A sorting device for multi-way recall results, comprising: The receiving module is configured to receive recommendation requests and obtain target user characteristics based on the recommendation requests; The determination module is configured to recall related item information of at least one related dimension based on the target user characteristics, and to determine the recall association weight and recall filtering weight of the related item information. The recall filtering weight is determined based on the time attribute information and the degree of association information in the related item information. The recall association weight refers to the association weight corresponding to each related dimension, and the recall filtering weight refers to the weight when filtering related item information under each related dimension. The acquisition module is configured to acquire target associated project information from the associated project information according to the recall screening weight; The sorting module is configured to sort the target associated project information based on the target associated project information and the recall associated weight to obtain an associated project sequence. The sorting includes: according to a preset sorting aggregation rule, sorting and aggregating the target associated project sequence composed of the target associated project information filtered under each associated dimension, combined with the target recall associated weight corresponding to each target associated project sequence, to obtain an associated project sequence.

8. A computing device comprising a memory, a processor, and computer instructions stored in the memory and executable on the processor, wherein the processor, when executing the computer instructions, performs the steps of the method according to any one of claims 1-5.

9. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-5.

10. A computer program product comprising a computer program, wherein, When the computer program is executed in a computer, it causes the computer to perform the steps of the method according to any one of claims 1-5.

Images