CN113781087B

CN113781087B - Recall method and device for recommended object, storage medium and electronic equipment

Info

Publication number: CN113781087B
Application number: CN202110127220.0A
Authority: CN
Inventors: 金均生; 朱发楠; 鄢雨; 刘明华
Original assignee: Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2021-01-29
Filing date: 2021-01-29
Publication date: 2024-10-22
Anticipated expiration: 2041-01-29
Also published as: CN113781087A

Abstract

The disclosure belongs to the technical field of computers, and relates to a recall method and device for a recommended object, a storage medium and electronic equipment. The method comprises the following steps: acquiring a historical object identifier, and determining a plurality of object identifiers to be recommended corresponding to the historical object identifier; wherein the plurality of object identifiers to be recommended are associated through a chain structure; calculating a plurality of scores to be recommended between the historical object identification and the plurality of object identifications to be recommended according to the chain structure; and determining recall object identifiers in the plurality of object identifiers to be recommended according to the plurality of scores to be recommended based on the chain structure, so as to recall according to the recall object identifiers. According to the method, on one hand, a chain structure is effectively combined with a calculation process, so that high-performance personalized recall service of billions of objects is realized, and calculation times and calculation time are effectively reduced; on the other hand, the performance requirements of high concurrency and low time delay of online service are met, and the recall performance in the offline process is also improved.

Description

Recall method and device for recommended object, storage medium and electronic equipment

Technical Field

The disclosure relates to the field of computer technology, and in particular, to a recall method of a recommended object, a recall device of the recommended object, a computer readable storage medium and an electronic device.

Background

With the popularity and rapid development of the mobile internet, online advertising has been developed. Advertisers of online advertisements can programmatically purchase media resources and automatically achieve accurate targeting of audience by utilizing algorithms and techniques, only delivering appropriate advertisements to paired people. In general, the method can be realized by two modes of a two-section recall technology based on collaborative filtering and an inner product model vector recall.

However, the two-stage recall technique based on "collaborative filtering" is not accurate enough and the diversity and discoverability of recall effects are poor. And the inner product model vector recall makes the cross-over specification not be effectively utilized, resulting in the recall effect being equally poor.

In view of this, there is a need in the art to develop a new recall method and apparatus for recommended objects.

It should be noted that the information disclosed in the above background section is only for enhancing understanding of the background of the present disclosure and thus may include information that does not constitute prior art known to those of ordinary skill in the art.

Disclosure of Invention

The present disclosure aims to provide a recall method of a recommended object, a recall device of a recommended object, a computer-readable storage medium and an electronic device, so as to overcome, at least to some extent, the technical problems of insufficient recall accuracy, poor recall effect and the like caused by the limitations of the related art.

Other features and advantages of the present disclosure will be apparent from the following detailed description, or may be learned in part by the practice of the disclosure.

According to a first aspect of an embodiment of the present invention, there is provided a recall method of a recommended object, the method including: acquiring a historical object identifier, and determining a plurality of object identifiers to be recommended corresponding to the historical object identifier; wherein the plurality of object identifiers to be recommended are associated through a chain structure;

Calculating a plurality of scores to be recommended between the historical object identification and the plurality of object identifications to be recommended according to the chain structure;

And determining recall object identifiers in the plurality of object identifiers to be recommended according to the plurality of scores to be recommended based on the chain structure, so as to recall according to the recall object identifiers.

In an exemplary embodiment of the present invention, the calculating a plurality of scores to be recommended between the historical object identification and the plurality of object identifications to be recommended according to the chain structure includes:

if the chain structure is a tree chain structure, determining a plurality of interval tree levels in the tree chain structure; wherein a next interval tree level of the plurality of interval tree levels is determined under a previous interval tree level;

And calculating a plurality of scores to be recommended between the historical object identification and a plurality of object identifications to be recommended in the plurality of interval tree levels.

If the chain structure is a continuous index structure, determining a plurality of interval index levels in the continuous index structure; wherein a next interval index level of the plurality of interval index levels is specified by a previous interval index level;

A plurality of scores to be recommended between the historical object identification and a plurality of objects to be recommended in the plurality of interval index levels is calculated.

if the historical object identifiers are a plurality of, calculating a plurality of historical object identifiers and a plurality of object identifiers to be recommended to obtain a score set;

And determining a plurality of scores to be recommended corresponding to a plurality of historical object identifications in the score set.

and inputting the historical object identification and the plurality of object identifications to be recommended into a pre-trained deep learning model according to the chain structure so that the deep learning model outputs a plurality of scores to be recommended.

In an exemplary embodiment of the present invention, the obtaining the historical object identifier includes:

Receiving an object recommendation request, wherein the object recommendation request carries terminal identification information;

and acquiring the historical object identification according to the terminal identification information.

In an exemplary embodiment of the present invention, the determining recall object identification from the plurality of to-be-recommended object identifications according to the plurality of to-be-recommended scores includes:

comparing the scores to be recommended to obtain a score comparison result;

and determining recall scores in the multiple to-be-recommended scores according to the score comparison results, and determining recall object identifications in the multiple to-be-recommended object identifications according to the recall scores.

According to a second aspect of an embodiment of the present invention, there is provided a recall device for a recommended object, the device including: the identification acquisition module is configured to acquire historical object identifications and determine a plurality of object identifications to be recommended corresponding to the historical object identifications; wherein the plurality of object identifiers to be recommended are associated through a chain structure;

the score calculating module is configured to calculate a plurality of scores to be recommended between the historical object identifications and the plurality of object identifications to be recommended according to the chain structure;

And the object determining module is configured to determine recall object identifiers in the plurality of object identifiers to be recommended according to the plurality of scores to be recommended based on the chain structure so as to recall according to the recall object identifiers.

According to a third aspect of an embodiment of the present invention, there is provided an electronic apparatus including: a processor and a memory; wherein the memory has stored thereon computer readable instructions which, when executed by the processor, implement the recall method of the recommended object of any of the exemplary embodiments described above.

According to a fourth aspect of embodiments of the present invention, there is provided a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the recall method of a recommended object in any of the above-described exemplary embodiments.

As can be seen from the above technical solutions, the recall method of the recommended object, the recall device of the recommended object, the computer storage medium and the electronic device in the exemplary embodiment of the present invention have at least the following advantages and positive effects:

In the method and the device provided by the exemplary embodiment of the disclosure, on one hand, the chain structure is effectively combined with the calculation process, so that the high-performance personalized recall service of the billion-magnitude object is realized, and the calculation times and the calculation time are effectively reduced; on the other hand, recall object identification is determined according to different chain structures, so that the performance requirements of high concurrency and low time delay of online service are met, and the recall performance in an offline process is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure. It will be apparent to those of ordinary skill in the art that the drawings in the following description are merely examples of the disclosure and that other drawings may be derived from them without undue effort.

FIG. 1 schematically illustrates a flowchart of a recall method of a recommended object in an exemplary embodiment of the present disclosure;

FIG. 2 schematically illustrates a flow diagram of a method of obtaining historical object identification in an exemplary embodiment of the present disclosure;

FIG. 3 schematically illustrates a flowchart of a method for determining recommendation scores in a tree chain structure in an exemplary embodiment of the present disclosure;

FIG. 4 schematically illustrates an effect diagram of a tree chain structure in an exemplary embodiment of the present disclosure;

fig. 5 schematically illustrates an effect diagram of a generation manner of a tree chain structure in an exemplary embodiment of the present disclosure;

FIG. 6 schematically illustrates a flowchart of a method for determining a score to be recommended by a continuous indexing structure in an exemplary embodiment of the present disclosure;

FIG. 7 schematically illustrates an effect diagram of a continuous index structure in an exemplary embodiment of the present disclosure;

FIG. 8 schematically illustrates an effect diagram of generating a tree hierarchy in cooperation with a deep learning model in an exemplary embodiment of the present disclosure;

FIG. 9 schematically illustrates a flow diagram of a method for determining a score to be recommended from a set of scores in an exemplary embodiment of the disclosure;

FIG. 10 schematically illustrates an effect diagram of calculating a set of scores in an exemplary embodiment of the present disclosure;

FIG. 11 schematically illustrates a flow diagram of a method of determining recall object identification in an exemplary embodiment of the present disclosure;

FIG. 12 schematically illustrates a flowchart of a recall method of a recommended object in an application scenario in an exemplary embodiment of the present disclosure;

FIG. 13 schematically illustrates a structure of a recall device of a recommended object in an exemplary embodiment of the present disclosure;

FIG. 14 schematically illustrates an electronic device for implementing a recall method of a recommended object in an exemplary embodiment of the disclosure;

FIG. 15 schematically illustrates a computer-readable storage medium for implementing a recall method of a recommended object in an exemplary embodiment of the disclosure.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments may be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the present disclosure. One skilled in the relevant art will recognize, however, that the aspects of the disclosure may be practiced without one or more of the specific details, or with other methods, components, devices, steps, etc. In other instances, well-known technical solutions have not been shown or described in detail to avoid obscuring aspects of the present disclosure.

The terms "a," "an," "the," and "said" are used in this specification to denote the presence of one or more elements/components/etc.; the terms "comprising" and "having" are intended to be inclusive and mean that there may be additional elements/components/etc. in addition to the listed elements/components/etc.; the terms "first" and "second" and the like are used merely as labels, and are not intended to limit the number of their objects.

Furthermore, the drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities.

With the popularity and rapid development of the mobile internet, online advertising has been developed. Online advertising refers to advertising of online media. Unlike traditional advertising, online advertising has formed a crowd-oriented, product-oriented, technical delivery model in the course of decades of short development.

Currently, a programmed advertisement platform performs advertisement transaction and management by using a technical means, an advertiser can programmatically purchase media resources, and an accurate target audience targeting is automatically realized by using an algorithm and a technology, so that only proper advertisements are put to a pair of people. However, with the dramatic growth of mobile user users and data, there is an increasing number of user points of interest, and how to use recommendation algorithms to deliver advertisements to paired people is a critical issue.

In an e-commerce environment, recommended advertisements are one of the most important forms of advertising for online advertising, which presents the user with goods that are most likely to be of interest to the user. The advertisement recall system is an important ring in the recommendation advertisement system, and the function of the advertisement recall system can be summarized by taking the historical behavior of the user as input, and recalling the commodity collection which is most likely to be interested by the user from the full commodity library.

Early recall systems traversed the full inventory, using simple algorithms, relying on less computing effort to recall a collection of target items from the inventory. With the development of the internet, the user quantity and commodity quantity are rapidly expanded, and the increasing speed exceeds the speed of computational effort progress, and a recall system for traversing the full commodity library cannot obtain a target commodity set in an acceptable time.

In the early stages of development of recommendation systems, in order to solve the contradiction between computing power and data volume, a two-stage recall technique based on "collaborative filtering" is generally used. The core idea of such methods is that "similar" users may be interested in "similar" merchandise.

The two-stage recall technique firstly clusters the commodities under the similarity label through various similarity calculation rules. During recall, the first stage obtains and screens user behavior (user historically browsed, focused and purchased goods), recalls some tags; and in the second stage, the commodities under the similarity label are output as a recall set.

In the two-stage recall process, the matching of commodities in the second stage is a core step, and the matching rule depends on the similarity of the commodities calculated offline. According to the triggering commodity and the label (one-stage result), K commodities with highest similarity are directly indexed. In general, the similarity of two commodities is characterized by the coincidence of respective user groups (browsing, clicking, focusing, etc. actions occur on the commodities). The online recall system based on the two-section matching does not need to traverse the full commodity library, the calculation complexity is only related to the richness of the user behaviors, the linear increase of the calculation cost along with the increase of the data quantity is avoided, the realization cost is low, the technology is mature, and the online recall system is widely applied in the industry.

With the development of interest modeling and indexing technology, the industry gradually transitions to inner product model vector recall to implement one-piece recall. At the index end, an increasingly perfect vector similarity recall technology provides an efficient guarantee for the application of the scheme; at the model end, the core idea is that by training the user interest model, the distance measurement (such as inner product distance and the like) between the user vector and the commodity vector produced by the model can represent the interest degree of the user on the commodity. When in online service, firstly, user behaviors are input into a user interest model, and a user vector is output; and then recalling the full library commodity by using the approximate vector, and recalling K commodities with highest similarity. Such a method enables for the first time a one-piece recall of a large candidate set.

The two-segment recall technique based on collaborative filtering is simpler to implement, but the overall accuracy is limited based on the rule-based similarity calculation and the two-segment recall mode of user-tag-commodity. In addition, since the overall recall concept is based on historical behavior to find similarities, recall results perform poorly in diversity and discoverability.

Inner product model vector recall because of the dependence on vector similarity, only inner product models can be used to measure the user's interest in the commodity, some more advanced model structures that can be used in the ranking stage, some user-commodity cross features, etc., cannot be used effectively. In addition, there is inconsistency in the goals of vector index construction and model optimization. The optimization objective of the product quantization index is to minimize approximation errors, while the optimization objective of vector recall is to maximize TopK recall. Resulting in sub-optimal final recall.

Aiming at the problems in the related art, the disclosure provides a recall method of a recommended object, which is applied to a terminal. FIG. 1 is a flow chart showing a recall method of a recommended object, and as shown in FIG. 1, the recall method of the recommended object at least comprises the following steps:

s110, acquiring a historical object identifier, and determining a plurality of object identifiers to be recommended corresponding to the historical object identifier; the object identifiers to be recommended are associated through a chain structure.

And S120, calculating a plurality of scores to be recommended between the historical object identifications and a plurality of object identifications to be recommended according to the chain structure.

And S130, determining recall object identifiers in the plurality of object identifiers to be recommended according to the plurality of scores to be recommended based on the chain structure so as to recall according to the recall object identifiers.

In the exemplary embodiment of the disclosure, on one hand, the chain structure is effectively combined with the calculation process, so that the high-performance personalized recall service of the billion-magnitude object is realized, and the calculation times and the calculation time are effectively reduced; on the other hand, recall object identification is determined according to different chain structures, so that the performance requirements of high concurrency and low time delay of online service are met, and the recall performance in an offline process is improved.

The following describes in detail the individual steps of the recall method of the recommended object.

In step S110, a history object identifier is obtained, and a plurality of object identifiers to be recommended corresponding to the history object identifier are determined; the object identifiers to be recommended are associated through a chain structure.

In an exemplary embodiment of the present disclosure, fig. 2 shows a flowchart of a method for obtaining a history object identifier, and as shown in fig. 2, the method at least includes the following steps: in step S210, an object recommendation request is received, the object recommendation request carrying terminal identification information.

The object recommendation request may be a request content such as an advertisement recommendation request, and the recommendation object request carries terminal identification information of a recommendation object. The terminal identification information may be identification information uniquely characterizing the terminal, or may be identification information uniquely characterizing a client used on the terminal, which is not particularly limited in this exemplary embodiment.

In step S220, a history object identification is acquired in accordance with the terminal identification information.

Further, the determining of the terminal identification information may be followed by searching the corresponding database for the historical object identification. The historical object identifier may be an identifier for characterizing a user historical browsing object, or may be an object identifier corresponding to other historical behaviors, which is not particularly limited in this exemplary embodiment.

In the present exemplary embodiment, the historical object identifier is obtained according to the terminal identifier information, so that the identifier information of the object of interest to the user history can be collected, and the prior information is provided for the accuracy of the subsequent determination of the recall object.

In step S120, a plurality of scores to be recommended between the historical object identifications and the plurality of object identifications to be recommended are calculated in a chain structure.

In an exemplary embodiment of the present disclosure, fig. 3 is a flowchart illustrating a method for determining a score to be recommended by a tree chain structure, and as shown in fig. 3, the method includes at least the following steps: in step S310, if the chain structure is a tree chain structure, determining a plurality of spaced tree levels in the tree chain structure; wherein a next interval tree level of the plurality of interval tree levels is determined under a previous interval tree level.

Fig. 4 shows the effect of the tree chain structure, as shown in fig. 4, taking the example of selecting an interval tree hierarchy structure across 2 levels of tree hierarchy, child nodes 2 and 3 are obtained from the root node 1, at this time, the root node 1 is the last interval tree hierarchy, and the child nodes 2 and 3 are temporarily ignored. But the hierarchy where the child nodes 4,5, 6 and 7 are located is obtained to the next level as an interval tree hierarchy, which is the next interval tree hierarchy. Obviously, the next interval tree level 4,5, 6 and 7 is below the level of the last interval tree level 1.

Since the tree-like chain structure satisfies the maximum logarithmic property, i.e. the father of an optimal TopK node of a certain layer must belong to the optimal TopK of the previous layer.

Therefore, the child node 6 is the optimal node of the present layer, and the sub-tree of the child node 6 must include the node of the optimal next interval tree level. With this loop, multiple alternate tree hierarchies in the tree chain structure can be reached.

In step S320, a plurality of scores to be recommended between the historical object identification and a plurality of object identifications to be recommended in a plurality of interval tree levels are calculated.

When the child node 6 is determined to be the optimal node of the interval tree hierarchy, a plurality of scores to be recommended between the historical object identifiers and the object identifiers to be recommended of each node in the interval tree hierarchy can be calculated, so that recall object identifiers are determined according to the plurality of scores to be recommended.

Fig. 5 shows an effect schematic diagram of the generation manner of the tree-like chain structure, and as shown in fig. 5, the tree-like chain structure has the maximum heap tree property and fully satisfies the property of the complete binary tree. The complete binary tree is a form of data organization and is also a form of binary tree. The complete binary tree has the smallest depth compared to the normal binary tree with the same number of nodes.

And, the leaf nodes of the tree chain structure represent all recommended objects, such as full library commodity, and the rest nodes represent abstract interest aggregate, and the top-down interest aggregate granularity is from thick to thin.

In the recall process, the top node of the tree chain structure, namely the root node, is selected first, and the object identifier to be recommended and the historical behavior identifier corresponding to the root node are calculated to obtain the corresponding score to be recommended. The to-be-recommended score may characterize a degree of interest of the user in identifying a corresponding object to the to-be-recommended object. Next, K nodes with the highest scores to be recommended are selected, and according to the maximum tree stacking property of the tree chain structure, objects of the K nodes with the highest matching degree with the user interests must exist in subtrees of the K nodes selected at this time, so that other nodes of the layer are pruned. Then, the sub-nodes of K nodes are selected in the next level or next interval tree hierarchy, and the calculation steps are repeated until the leaf level is reached.

Obviously, when calculating the plurality of scores to be recommended for the tree chain structure, the scores in the tree chain structure can be calculated layer by layer, or the scores in a plurality of interval tree levels selected by cross-level calculation, and the exemplary embodiment is not particularly limited to this.

According to the maximum tree stacking property of the tree chain structure, for any optimal node of the nth layer, the father node of the node is necessarily the optimal node at the n-1 layer. Therefore, as long as the optimal node of each layer is selected in the top-down searching process, the optimal node of the next layer is necessarily a child node of the selected node of the layer, and the final recall object identification can be determined at the leaf layer of the tree-shaped chain structure.

For example, when the number of the commodities in the whole library is N and the number of recalls is K, the score to be recommended of each layer of the tree chain structure is calculated, the calculated time complexity is O (N), and the calculated time complexity isI.e. number of recalls 2 model calculation times (depth of tree). In practical application, N usually reaches a hundred million level, K is in a hundred magnitude, and 1 hundred million and 200 are taken as examples respectively, and the N is required to score 1 hundred million commodities through a tree chain structure, and only 7400 times of calculation are required.

Further, when searching the interval tree hierarchy of the tree chain structure, the calculated time can be changed from 4 to 2 (the root node does not need to be calculated actually), and the number of the input nodes of the word is changed from 2 to 4. That is, the m-layer computation reduces the computation times by m times, the single computation scale is increased by m times, the complexity of the overall computation is unchanged, and the efficiency is greatly improved.

In the present exemplary embodiment, by calculating the score to be recommended in the interval tree hierarchy in the tree chain structure, the data amount of each calculation can be increased and the number of times of calculation can be reduced on the premise of keeping the complexity unchanged, so that the calculation efficiency is greatly improved.

FIG. 6 is a flow chart of a method for determining scores to be recommended by a continuous index structure, and the method at least comprises the following steps as shown in FIG. 6: in step S610, if the chain structure is a continuous index structure, determining a plurality of interval index levels in the continuous index structure; wherein a next interval index level of the plurality of interval index levels is specified by a previous interval index level.

In the off-line calculation process, the calculation mode of the tree chain structure shown in fig. 3 can be adopted. In order to better improve recall efficiency in the online calculation process, a continuous index structure can be adopted for calculation.

FIG. 7 is a schematic diagram showing the effect of the continuous index structure, and as shown in FIG. 7, the candidate nodes of each node of the tree chain structure are clustered into continuous memory space, i.e. the interval index level. At the same time, an index exists between each node and its interval index level, the index is also stored in a section of continuous memory space, and the sequence number of each index in the memory space is the sequence number of the node.

The continuous memory identification data are stored in a section of continuous control, and the address of any data can be directly found according to the index of the data.

Specifically, starting from the node 1, candidate nodes 4, 5, 6 and 7 are designated according to the index of the position "1", and the identification of the object to be recommended of the candidate nodes is calculated to obtain the optimal node 6. Further, candidate nodes 24, 25, 26 and 27 are determined from the index of location "6" for subsequent calculation until the final recall object identification is determined.

In step S620, a plurality of to-be-recommended scores between the historical object identification and a plurality of to-be-recommended objects in a plurality of interval index hierarchies are calculated.

When determining that the optimal node is the candidate node 6, the historical object identifier and the object identifiers to be recommended of the candidate nodes 4, 5, 6 and 7 may be calculated to obtain corresponding scores to be recommended, so as to determine that the optimal node is the candidate node 6 according to the scores to be recommended.

In the present exemplary embodiment, the score to be recommended is calculated by adopting a continuous index structure, so that layer-by-layer traversal of the tree chain structure is avoided, and the operation efficiency in the online calculation process is improved.

The manner of calculating the score to be recommended in step S320 and step S620 may be implemented by a deep learning model.

In an alternative embodiment, the historical object identification and the plurality of object identifications to be recommended are input into a pre-trained deep learning model according to a chain structure, so that the deep learning model outputs a plurality of scores to be recommended.

Specifically, a deep learning model may be used to fit the classification samples, pulling the deep learning to approximate the maximum heap tree property. The classification samples contain positive samples (objects of interest) and negative samples (objects of no interest), which can be randomly sampled at each layer.

The generation principle of the tree hierarchy structure and the deep learning model is to ensure that the calculated scores of the deep learning model on the nodes of the tree hierarchy structure meet the maximum tree piling property.

Fig. 8 shows an effect schematic diagram of a tree hierarchy structure generated in cooperation with a deep learning model, as shown in fig. 8, the node 18 is a commodity clicked by a user in a piece of sample data, namely a positive sample, and according to the maximum heap tree property, a parent node of the positive sample, a parent node of the parent sample, and the like are positive samples, namely nodes 9, 4, 2 and 1.

For the deep learning model, the weights of each node of the deep learning model may be adjusted such that positive sample nodes 18, 9, 4, 2, and 1 get the highest scores among all nodes of the layer at which the nodes are located, and the scoring scores of the negative samples are the lowest.

Since the deep learning model only focuses on sample fitting, the deep learning model may be a deep interest Network (DEEP INTEREST Network, abbreviated as DIN) or other deep learning models, and the present exemplary embodiment is not limited thereto. Therefore, the upper limit of the optimization of this calculation method is also described as extremely high.

In the calculation, a plurality of batches of calculation may be performed using the deep learning model, in addition to the calculation according to step S320 and step S620.

In an alternative embodiment, fig. 9 shows a flow chart of a method for determining a score to be recommended according to a score set, and as shown in fig. 9, the method at least includes the following steps: in step S910, if the historical object identifiers are plural, a score set is obtained by calculating the plurality of historical object identifiers and the plurality of object identifiers to be recommended.

When multiple object recommendation requests are received, multiple historical object identifications may be obtained. Further, the calculation processes between the plurality of historical object identifications and the object to be recommended can be combined to obtain a corresponding score set.

Fig. 10 shows an effect of calculating a score set, and as shown in fig. 10, the three historical object identifications and the multiple to-be-recommended object identifications are combined into the same input calculation, and the input calculation is taken out from the to-be-processed queue and input into the deep learning model, so that the deep learning model outputs a score set.

In step S920, a plurality of scores to be recommended corresponding to the plurality of history object identifications are determined in the score set.

Further, the score sets calculated by the deep learning model are distributed, namely, scores to be recommended corresponding to the plurality of history identification objects are distributed to different input parties respectively.

The manner in which the score set is calculated and the score to be recommended is determined from the score set may be implemented based on an Nvidia graphics computing card (graphics processor Graphics Processing Unit, GPU for short).

GPUs have extremely powerful parallel computing capabilities that make computation time consuming to compute K values over a range unchanged. Aiming at the parallel computing characteristics of the GPU, an optimization strategy for combining and processing the computing tasks of multiple batches can be realized.

In the present exemplary embodiment, the score sets are obtained through parallel computation, so as to determine a plurality of scores to be recommended, thereby reducing the computation times, improving the throughput of GPU computation, reducing the time consumption of query, avoiding the cost loss of frequent GPU interaction, and improving the GPU utilization rate.

In step S130, based on the chain structure, recall object identifiers are determined from the plurality of to-be-recommended object identifiers according to the plurality of to-be-recommended scores, so as to recall according to the recall object identifiers.

In an exemplary embodiment of the present disclosure, when determining a plurality of scores to be recommended layer by layer according to the chain structure, this calculation process may be cycled to determine recall object identification among the plurality of object identifications to be recommended.

In an alternative embodiment, FIG. 11 shows a flow chart of a method of determining recall object identification, as shown in FIG. 11, the method comprising at least the steps of: in step S1110, a plurality of scores to be recommended are compared to obtain a score comparison result.

And comparing the scores to be recommended, and sorting the scores to be recommended in a relatively large and small mode. Specifically, the sorting manner may be from large to small or from small to large, which is not particularly limited in the present exemplary embodiment.

In step S1120, a recall score is determined from the plurality of to-be-recommended scores according to the score comparison result, and a recall object identification is determined from the plurality of to-be-recommended object identifications according to the recall score.

According to actual requirements, the recall score with the highest score can be determined from the multiple to-be-recommended scores, and the recall score with the same recall number can be selected according to the requirement of the recall number, so that the exemplary embodiment is not particularly limited.

And after determining the recall score, determining the object identifier to be recommended corresponding to the recall score as a recall object identifier.

In the present exemplary embodiment, the recall object identifier may be determined in the object identifier to be recommended by a comparison manner, and the determination manner is simple and accurate, and is applicable to the requirements of multiple recall scenes.

After determining the recall object identification, the corresponding recall object can be determined to carry out recall processing according to the recall object identification.

The recall method of the recommended object in the embodiment of the present disclosure is described in detail below in connection with an application scenario.

Fig. 12 is a flowchart illustrating a recall method of a recommended object in an application scenario, and in step S1210, an advertisement recommendation request is acquired, where the advertisement recommendation request carries client identification information of the recommended object, that is, terminal identification information, as shown in fig. 12.

In step S1220, the historical object identifier corresponding to the historical behavior is searched in the database according to the client identifier information.

In step S1230, the nodes of the tree chain structure are selected top-down.

In step S1240, the historical object identification of the user' S historical behavior and the object identification to be recommended of the selected node are input into the deep learning model.

In step S1250, the output result of the model is the score of the tree node, and the TopK nodes are selected according to the score. Specifically, the output result of the deep learning model is the score to be recommended of the nodes of the tree chain structure, and TopK nodes are selected according to the score to be recommended.

In step S1260, the left and right child nodes of TopK nodes are selected as input data estimated by the next deep learning model.

In step S1270, steps S1240 to S1260 are repeated until the leaf nodes of the tree chain structure are traversed and the leaf node scoring is completed.

In step S1280, the results of recommended commodities for which TopK leaf nodes are targets are returned.

The tree chain structure organizes commodities in a tree structure, and optimal tree nodes are searched layer by layer in the inquiring process until leaf layers are reached. The deep learning model takes the identification of the object to be recommended of the tree node and the identification of the historical object of the user behavior as inputs, and inputs the interests and the scores of the user for each node. The tree chain structure and the deep learning model work cooperatively, and have strong consistency, namely, the tree chain structure and the deep learning model are generated by training the same data set, correct each other in training, and the version consistency is also required to be kept in the working process.

In the application scene of the method, on one hand, the chain structure is effectively combined with the calculation process, so that the high-performance personalized recall service of the billion-level objects is realized, and the calculation times and the calculation time are effectively reduced; on the other hand, recall object identification is determined according to different chain structures, so that the performance requirements of high concurrency and low time delay of online service are met, and the recall performance in an offline process is improved.

In addition, in the exemplary embodiment of the disclosure, a recall device of the recommended object is also provided. Fig. 13 shows a schematic structural diagram of a recall device of a recommended object, and as shown in fig. 13, a recall device 1300 of a recommended object may include: an identification acquisition module 1310, a score calculation module 1320, and an object determination module 1330. Wherein:

An identifier acquisition module 1310 configured to determine a historical object identifier and determine a plurality of object identifiers to be recommended corresponding to the historical object identifier; wherein, a plurality of object identifications to be recommended are associated through a chain structure; a score calculation module 1320 configured to calculate a plurality of scores to be recommended between the historical object identifications and the plurality of object identifications to be recommended in a chain structure; the object determination module 1330 is configured to determine, based on the chain structure, recall object identifications among the plurality of to-be-recommended object identifications according to the plurality of to-be-recommended scores, so as to recall according to the recall object identifications.

The specific details of the recall device 1300 of the recommended object are described in detail in the recall method of the corresponding recommended object, and thus are not described herein.

It should be noted that although several modules or units of recall device 1300 of a recommended object are mentioned in the detailed description above, such partitioning is not mandatory. Indeed, the features and functionality of two or more modules or units described above may be embodied in one module or unit in accordance with embodiments of the present disclosure. Conversely, the features and functions of one module or unit described above may be further divided into a plurality of modules or units to be embodied.

In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

An electronic device 1400 according to such an embodiment of the invention is described below with reference to fig. 14. The electronic device 1400 shown in fig. 14 is merely an example and should not be construed as limiting the functionality and scope of use of embodiments of the present invention.

As shown in fig. 14, the electronic device 1400 is embodied in the form of a general purpose computing device. Components of electronic device 1400 may include, but are not limited to: the at least one processing unit 1410, the at least one memory unit 1420, a bus 1430 connecting the different system components (including the memory unit 1420 and the processing unit 1410), and a display unit 1440.

Wherein the storage unit stores program code that is executable by the processing unit 1410 such that the processing unit 1410 performs steps according to various exemplary embodiments of the present invention described in the above section of the "exemplary method" of the present specification.

The memory unit 1420 may include readable media in the form of volatile memory units, such as Random Access Memory (RAM) 1421 and/or cache memory 1422, and may further include Read Only Memory (ROM) 1423.

The memory unit 1420 may also include a program/utility 1424 having a set (at least one) of program modules 1425, such program modules 1425 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment.

Bus 1430 may be a local bus representing one or more of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or using any of a variety of bus architectures.

The electronic device 1400 may also communicate with one or more external devices 1600 (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 1400, and/or with any device (e.g., router, modem, etc.) that enables the electronic device 1400 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 1450. Also, electronic device 1400 may communicate with one or more networks such as a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet, through network adapter 1460. As shown, the network adapter 1460 communicates with other modules of the electronic device 1400 via the bus 1430. It should be appreciated that although not shown, other hardware and/or software modules may be used in connection with electronic device 1400, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

From the above description of embodiments, those skilled in the art will readily appreciate that the example embodiments described herein may be implemented in software, or in combination with the necessary hardware. Thus, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.) or on a network, and includes several instructions to cause a computing device (may be a personal computer, a server, a terminal device, or a network device, etc.) to perform the method according to the embodiments of the present disclosure.

In an exemplary embodiment of the present disclosure, a computer-readable storage medium having stored thereon a program product capable of implementing the method described above in the present specification is also provided. In some possible embodiments, the various aspects of the invention may also be implemented in the form of a program product comprising program code for causing a terminal device to carry out the steps according to the various exemplary embodiments of the invention as described in the "exemplary methods" section of this specification, when said program product is run on the terminal device.

Referring to fig. 15, a program product 1500 for implementing the above-described method according to an embodiment of the present invention is described, which may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The computer readable signal medium may include a data signal propagated in baseband or as part of a carrier wave with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

Claims

1. A recall method of a recommended object, the method comprising:

Acquiring a historical object identifier, and determining a plurality of object identifiers to be recommended corresponding to the historical object identifier; wherein the plurality of object identifiers to be recommended are associated through a chain structure;

determining recall object identifiers in the plurality of object identifiers to be recommended according to the plurality of scores to be recommended based on the chain structure so as to recall according to the recall object identifiers;

the calculating, according to the chain structure, a plurality of scores to be recommended between the historical object identifier and the plurality of object identifiers to be recommended includes:

And calculating a plurality of scores to be recommended between the historical object identifications and the plurality of object identifications to be recommended in the plurality of interval tree levels.

2. The recall method of recommended objects according to claim 1, wherein the calculating a plurality of scores to be recommended between the historical object identification and the plurality of object identifications to be recommended according to the chain structure further comprises:

A plurality of scores to be recommended between the historical object identification and the plurality of objects to be recommended in the plurality of interval index levels is calculated.

3. The recall method of recommended objects according to claim 1, wherein the calculating a plurality of scores to be recommended between the historical object identification and the plurality of object identifications to be recommended according to the chain structure further comprises:

4. The recall method of recommended objects according to claim 1, wherein the calculating a plurality of scores to be recommended between the historical object identification and the plurality of object identifications to be recommended according to the chain structure further comprises:

5. The recall method of recommended objects of claim 1 wherein the obtaining historical object identification comprises:

6. The recall method of a recommended object according to any one of claims 1 to 5, wherein determining a recall object identification among the plurality of to-be-recommended object identifications according to the plurality of to-be-recommended scores comprises:

comparing the scores to be recommended to obtain a score comparison result;

7. A recall device for a recommended object, comprising:

The identification acquisition module is configured to acquire historical object identifications and determine a plurality of object identifications to be recommended corresponding to the historical object identifications; wherein the plurality of object identifiers to be recommended are associated through a chain structure;

The object determining module is configured to determine recall object identifiers in the plurality of object identifiers to be recommended according to the plurality of scores to be recommended based on the chain structure so as to recall according to the recall object identifiers;

the calculating, according to the chain structure, a plurality of scores to be recommended between the historical object identifier and the plurality of object identifiers to be recommended includes: if the chain structure is a tree chain structure, determining a plurality of interval tree levels in the tree chain structure; wherein a next interval tree level of the plurality of interval tree levels is determined under a previous interval tree level; and calculating a plurality of scores to be recommended between the historical object identifications and the plurality of object identifications to be recommended in the plurality of interval tree levels.

8. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the recall method of a recommended object according to any one of claims 1-6.

9. An electronic device, comprising:

A processor;

A memory for storing executable instructions of the processor;

Wherein the processor is configured to perform the recall method of the recommended object of any one of claims 1-6 via execution of the executable instructions.