CN115033801A

CN115033801A - Article recommendation method, model training method and electronic equipment

Info

Publication number: CN115033801A
Application number: CN202210953005.0A
Authority: CN
Inventors: 连德富; 陈恩红; 冯超; 黎武超
Original assignee: University of Science and Technology of China USTC
Current assignee: University of Science and Technology of China USTC
Priority date: 2022-08-10
Filing date: 2022-08-10
Publication date: 2022-09-09
Anticipated expiration: 2042-08-10
Also published as: CN115033801B

Abstract

The invention provides an article recommendation method, a model training method and electronic equipment. The method comprises the following steps: acquiring user characteristic information of a candidate item set and a target user; performing balanced clustering analysis on the candidate article sets based on a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees, wherein each non-leaf node of each K-ary tree corresponds to one article set, each article set corresponds to one path sequence, the path sequence corresponding to each article set represents a path from a root node of the K-ary tree to each non-leaf node corresponding to each article set, and each leaf node of each K-ary tree corresponds to one candidate article; determining a target path from a root node to a leaf node according to user characteristic information and a path sequence corresponding to a non-leaf node for each K-ary tree, and taking candidate articles corresponding to the leaf nodes on the target path as target candidate articles; and determining the target recommended item according to the N target candidate items.

Description

Article recommendation method, model training method and electronic equipment

Technical Field

The invention relates to the technical field of computer technology and machine learning, in particular to an article recommendation method, a model training method and electronic equipment.

Background

With the continuous development of electronic technology and network technology, more and more users enjoy online shopping. At present, a service platform recommends an item to a user, which is usually based on a recommendation system, inputs user characteristic information into the recommendation system, calculates the similarity between the user characteristic information and massive item characteristic information in a database based on a similarity measurement function, converts a recommendation problem into a nearest neighbor search problem, selects a small number (for example, hundreds) of candidate item objects with high relevance to the user by using an index structure of the nearest neighbor search problem, sorts the candidate items, and finally displays a recommendation result to the user.

However, the item recommendation method is complex in calculation degree, the user characteristic information representation process and the item recommendation process are independent from each other, interaction of information of the user characteristic information representation process and the item recommendation process cannot be supported, and due to the fact that the similarity function expression capability is limited, the user's liking degree of the item is difficult to accurately depict, and accuracy of a recommendation result is affected.

Disclosure of Invention

In view of the above problems, the present invention provides an article recommendation method, a model training method, and an electronic device.

One aspect of the present invention provides an item recommendation method, including: acquiring user characteristic information of a candidate item set and a target user; performing balanced clustering analysis on candidate item sets based on a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees, wherein each non-leaf node of each K-ary tree corresponds to one item set, each item set corresponds to one path sequence, the path sequence corresponding to each item set represents a path from a root node of the K-ary tree to each non-leaf node corresponding to each item set, each leaf node of each K-ary tree corresponds to one candidate item, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2; determining a target path from a root node to a leaf node according to user characteristic information and a path sequence corresponding to an article set aiming at each K-ary tree, and taking a candidate article corresponding to the leaf node on the target path as a target candidate article; and determining a target recommended item according to N target candidate items, wherein N is a positive integer greater than or equal to 1.

Optionally, determining a target path from the root node to the leaf node according to the user feature information and the path sequence corresponding to the item set, and taking a candidate item corresponding to the leaf node on the target path as a target candidate item, including: extracting the characteristics of the user characteristic information to obtain a user characteristic vector; predicting the direction representation of a next layer node of the current layer node by the path sequence corresponding to the article set of the current layer node according to the user feature vector and the path sequence corresponding to the article set based on the sequence from the root node to the leaf node, and generating the path sequence corresponding to the article set of the next layer node; under the condition that the next-layer node is a non-leaf node, taking the next-layer node as a new current-layer node, repeatedly executing the direction representation of the next-layer node of the new current-layer node predicted by the path sequence corresponding to the article set of the new current-layer node, and generating a path sequence corresponding to the article set of the next-layer node of the new current-layer node; and under the condition that the next layer of nodes are leaf nodes, generating a target path from the root node to the leaf nodes, and determining candidate articles corresponding to the leaf nodes on the target path as target candidate articles.

Optionally, determining a target recommended item according to the N target candidate items includes: taking a union set of the N target candidate articles to generate a target candidate article set; determining a recommendation probability value of each target candidate item in the target candidate item set; based on a preset sorting rule, sorting the recommendation probability value of each target candidate item to obtain a sorting result of the recommendation probability values of the target candidate items; and determining the target recommended item according to the sequencing result of the recommendation probability values of the target candidate items.

Optionally, based on a hierarchical balanced clustering algorithm with randomness, performing balanced clustering analysis on the candidate item set to construct M K-ary trees, including: performing feature extraction on candidate articles in the candidate article set aiming at each K-ary tree to obtain a plurality of candidate article characterization vectors, wherein the candidate article characterization vectors are used as root nodes of the K-ary tree; determining K clustering centers of the candidate item characterization vectors; performing first balanced clustering analysis on the candidate item characterization vectors of the root node based on the K clustering centers and the candidate item characterization vectors to obtain K clusters of the root node, wherein each cluster comprises a plurality of candidate item characterization vectors with the same quantity; obtaining K clustering results of the root nodes according to the K clusters of the root nodes; taking the K clustering results of the root node as K non-leaf nodes of a first layer of the root node; and performing second balanced clustering analysis on each non-leaf node until each obtained clustering result comprises a candidate article characterization vector, and taking each clustering result comprising the candidate article characterization vector as a leaf node to obtain the K-ary tree.

Another aspect of the present invention provides an article recommendation model training method, including: acquiring a training sample, wherein the training sample comprises sample user characteristic information and a historical path sequence of a label sample article; training an item recommendation model to be trained based on a training sample to obtain a trained item recommendation model, wherein a historical path sequence of a label sample item is obtained through a path sequence of a candidate item corresponding to the label sample item in M constructed K-ary trees, the M constructed K-ary trees are obtained by performing balanced clustering analysis on a candidate item set based on a hierarchical balanced clustering algorithm with randomness, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2.

Optionally, training the recommendation model for the item to be trained based on the training samples includes: performing feature extraction on sample user feature information in a training sample aiming at each K-ary tree to obtain a sample user feature vector; determining an objective function value according to the sample user characteristic vector and the historical path sequence of the label sample article; and adjusting the model parameters of the recommendation model of the object to be trained based on the objective function value.

Optionally, the historical path sequence of the labeled sample item includes a given historical path sequence for each layer of the labeled sample item in the K-ary tree.

Optionally, determining an objective function value according to the sample user feature vector and the historical path sequence of the label sample item, includes: predicting the probability value of the path direction representation of the next layer of each layer in the K-ary tree corresponding to the sample user feature vector according to the sample user feature vector and the given historical path sequence of each layer in the K-ary tree to obtain the probability value of the path direction representation of each layer in the K-ary tree; and determining an objective function value according to the probability value represented by the path direction of each layer.

Optionally, the item recommendation model comprises an encoder and a decoder.

Optionally, the Decoder is a transform Decoder structure.

Another aspect of the present invention also provides an article recommendation apparatus, including: the first acquisition module is used for acquiring the candidate item set and the user characteristic information of the target user; the building module is used for carrying out balanced clustering analysis on the candidate item sets based on a hierarchical balanced clustering algorithm with randomness to build M K-ary trees, wherein each non-leaf node of each K-ary tree corresponds to one item set, each item set corresponds to one path sequence, the path sequence corresponding to each item set represents a path from a root node of the K-ary tree to each non-leaf node corresponding to each item set, each leaf node of each K-ary tree corresponds to one candidate item, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2; the first determining module is used for determining a target path from the root node to the leaf node according to the user characteristic information and the path sequence corresponding to the item set aiming at each K-ary tree, and taking the candidate item corresponding to the leaf node on the target path as a target candidate item; and the second determination module is used for determining the target recommended item according to the N target candidate items, wherein N is a positive integer greater than or equal to 1.

Another aspect of the present invention further provides an article recommendation model training apparatus, including: the second acquisition module is used for acquiring a training sample, and the training sample comprises sample user characteristic information and a historical path sequence of a label sample article; the training module is used for training the item recommendation model to be trained on the basis of the training samples to obtain the trained item recommendation model, wherein the historical path sequence of the label sample item is obtained through the path sequence of the candidate item corresponding to the label sample item in M constructed K-ary trees, the M constructed K-ary trees are obtained by carrying out balanced clustering analysis on the candidate item set on the basis of a hierarchical balanced clustering algorithm with randomness, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2.

Another aspect of the present invention also provides an electronic device, including: one or more processors; a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the item recommendation method and the item recommendation model training method described above.

Yet another aspect of the present invention provides a computer-readable storage medium having stored thereon executable instructions, which when executed by a processor, cause the processor to perform the above item recommendation method and item recommendation model training method.

Another aspect of the present invention also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the computer program implements the item recommendation method and the item recommendation model training method described above.

The method comprises the steps of establishing M K-ary trees, determining a target path from a root node to a leaf node based on a path sequence and user characteristic information of an article set corresponding to each non-leaf node of the M K-ary trees, and determining candidate articles corresponding to the leaf nodes on the target path as target recommended articles. The candidate item set is subjected to clustering analysis through a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees, so that the diversity and randomness of the M K-ary trees are guaranteed, the path sequence of the target candidate item is determined by the path sequence index of the item set of non-leaf nodes, and the information of the node at the upper layer of the K-ary tree structure is considered in the process of determining the target candidate item, so that the item at the division edge can be recommended to a user, and the item recommendation accuracy is improved.

Drawings

The foregoing and other objects, features and advantages of the invention will be apparent from the following description of embodiments of the invention, which proceeds with reference to the accompanying drawings, in which:

FIG. 1 is a diagram schematically illustrating an application scenario of an item recommendation method, an item recommendation model training method and an apparatus according to an embodiment of the present invention;

FIG. 2 schematically shows a flow diagram of an item recommendation method according to an embodiment of the invention;

FIG. 3(a) is a schematic diagram that illustrates a tree structure and path sequence of a binary tree according to an embodiment of the invention;

FIG. 3(b) is a diagram schematically illustrating a tree structure and a path sequence of a binary tree according to another embodiment of the present invention;

FIG. 4 is a flow chart of a method for determining a candidate item corresponding to a leaf node as a target candidate item according to an embodiment of the invention;

FIG. 5 schematically illustrates a flow diagram of an item recommendation model training method according to an embodiment of the invention;

fig. 6 is a block diagram schematically showing the configuration of an item recommendation apparatus according to an embodiment of the present invention;

FIG. 7 is a block diagram schematically illustrating the structure of an item recommendation model training apparatus according to an embodiment of the present invention;

fig. 8 schematically shows a block diagram of an electronic device adapted to implement an item recommendation method and an item recommendation model training method according to an embodiment of the present invention.

Detailed Description

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. It is to be understood that such description is merely illustrative and not intended to limit the scope of the present invention. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the invention. It may be evident, however, that one or more embodiments may be practiced without these specific details. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The terms "comprises," "comprising," and the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.

All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It is noted that the terms used herein should be interpreted as having a meaning that is consistent with the context of this specification and should not be interpreted in an idealized or overly formal sense.

Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B and C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).

A binary tree can be used for representing the item set by utilizing a depth model based on a tree structure, each middle node represents the clustering center of all descendant leaf nodes of the middle node, each leaf node represents an item, each layer of nodes are scored from the top to the bottom of the root node of the tree, and finally, the items corresponding to the part of leaf nodes with the highest scores in the leaf nodes are recommended.

The inventor finds that in the process of implementing the concept of the invention, the recommendation mode based on the tree structure is limited by dividing the article set layer by layer, and articles at the dividing edge are difficult to retrieve. In the retrieval process, historical nodes positioned in front of a current layer are not considered, only the nodes of the current layer are considered, information loss is easy to cause, the representation of each node is stored in a tree structure, the space complexity is equal to the quantity of the article sets, and the memory consumption is large.

In view of this, an embodiment of the present invention provides an item recommendation method, including: acquiring user characteristic information of a candidate item set and a target user; performing balanced clustering analysis on candidate item sets based on a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees, wherein each non-leaf node of each K-ary tree corresponds to one item set, each item set corresponds to one path sequence, the path sequence corresponding to each item set represents a path from a root node of the K-ary tree to each non-leaf node corresponding to each item set, each leaf node of each K-ary tree corresponds to one candidate item, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2; for each K-ary tree, determining a target path from a root node to a leaf node according to the user characteristic information and a path sequence corresponding to the article set, and taking a candidate article corresponding to the leaf node on the target path as a target candidate article; and determining a target recommended item according to N target candidate items, wherein N is a positive integer greater than or equal to 1.

In the technical scheme of the invention, the collection, storage, use, processing, transmission, provision, disclosure, application and other processing of the personal information of the related user are all in accordance with the regulations of related laws and regulations, necessary confidentiality measures are taken, and the customs of the public order is not violated.

In the technical scheme of the invention, the data acquisition, collection, storage, use, processing, transmission, provision, disclosure, application and other processing are all in accordance with the regulations of relevant laws and regulations, necessary security measures are taken, and the public order and good custom are not violated.

Fig. 1 schematically shows an application scenario diagram of an item recommendation method, an item recommendation model training method and an apparatus according to an embodiment of the present invention.

As shown in fig. 1, the application scenario 100 according to this embodiment may include

terminal devices

101, 102, 103, a network 104 and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

The user may use the

terminal devices

101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. The

terminal devices

101, 102, 103 may have installed thereon various communication client applications, such as shopping-like applications, web browser applications, search-like applications, instant messaging tools, mailbox clients, social platform software, etc. (by way of example only).

The

terminal devices

101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.

The server 105 may be a server providing various services, such as a background management server (for example only) providing support for websites browsed by users using the

terminal devices

101, 102, 103. The background management server may analyze and perform other processing on the received data such as the user request, and feed back a processing result (e.g., a webpage, information, or data obtained or generated according to the user request) to the terminal device.

It should be noted that the item recommendation method and the item recommendation model training method provided in the embodiments of the present invention may be generally executed by the server 105. Accordingly, the item recommendation device and the item recommendation model training device provided by the embodiment of the present invention may be generally disposed in the server 105. The item recommendation method and the item recommendation model training method provided by the embodiment of the present invention may also be executed by a server or a server cluster that is different from the server 105 and can communicate with the

terminal devices

101, 102, 103 and/or the server 105. Correspondingly, the item recommendation apparatus and the item recommendation model training apparatus provided in the embodiments of the present invention may also be disposed in a server or a server cluster that is different from the server 105 and can communicate with the

terminal devices

101, 102, and 103 and/or the server 105.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

Fig. 2 schematically shows a flow chart of an item recommendation method according to an embodiment of the invention.

As shown in FIG. 2, the method 200 includes operations S210 to S240.

In operation S210, user characteristic information of the candidate item set and the target user is acquired.

In operation S220, based on a hierarchical balanced clustering algorithm with randomness, performing balanced clustering analysis on candidate item sets to construct M K-ary trees, where each non-leaf node of each K-ary tree corresponds to an item set, each item set corresponds to a path sequence, the path sequence corresponding to each item set characterizes a path from a root node of the K-ary tree to each non-leaf node corresponding to each item set, each leaf node of each K-ary tree corresponds to a candidate item, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2.

In operation S230, for each K-ary tree, a target path from the root node to a leaf node is determined according to the user feature information and the path sequence corresponding to the item set, and a candidate item corresponding to the leaf node on the target path is taken as a target candidate item.

In operation S240, a target recommended item is determined according to N target candidate items, where N is a positive integer greater than or equal to 1.

According to an embodiment of the present invention, the candidate item set may be a set of all items obtained from all accessible shopping platforms, or may be a set of all items provided by a plurality of target shopping platforms, which may be recommended shopping platforms for the target user, or may be a set of items provided by a shopping platform currently selected by the target user. The candidate items in the candidate item set may include items purchased by the user or items not purchased by the user.

According to the embodiment of the invention, the target user can be a user recommending articles for the user, and can be any user who purchased or clicked to browse articles on an online shopping platform in the past. The user characteristic information of the target user can be basic information of the target user or item information of an item purchased or clicked by the target user on the shopping platform. The basic information of the target user can comprise information such as age, gender, city, account number of a shopping platform and the like.

According to an embodiment of the present invention, randomness may refer to uniformly and randomly distributing commodities in a candidate item set to different clustering centers when the candidate item set is clustered. Layering may refer to clustering the items in the candidate set of items in order from a root node to a leaf node in the K-ary tree. The balanced clustering algorithm can realize that the items in the candidate item set are uniformly distributed to each different clustering center, namely, the item set of each layer of nodes is uniformly distributed to the next layer of nodes of the nodes, thereby constructing the K-ary tree.

According to the embodiment of the invention, M K-ary trees are constructed based on a balanced clustering algorithm, M and K are positive integers greater than or equal to 2, and the specific values of M and K can be determined according to actual needs.

According to the embodiment of the invention, each K-ary tree in the M K-ary trees can be obtained by performing balanced clustering analysis on the candidate item set through a balanced clustering algorithm. For each K-ary tree, the K-ary tree may be a balanced K-ary tree, where each node except leaf nodes has K branches.

According to embodiments of the invention, the K-ary tree may include multiple levels from a top level to a bottom level, each of which may include one or more nodes. The top level of the K-ary tree may include only the root node, the bottom level nodes of the K-ary tree have no children, and the bottom level nodes of the K-ary tree may be referred to as leaf nodes. In addition, nodes in the top to middle layers of the K-ary tree, i.e., the non-bottom layers, have child nodes, and the nodes of the non-bottom layers of the K-ary tree may be referred to as non-leaf nodes.

According to an embodiment of the present invention, each node in the K-ary tree except for leaf nodes has K child nodes, i.e., K branches. The K branches in the K-ary tree may be K different directions, each direction includes a direction token, and the direction tokens for the K directions may be {0,1,2, …, K-1 }.

According to the embodiment of the invention, each non-leaf node in the K-ary tree corresponds to one article set, each article set corresponds to one path sequence, and the path sequence corresponding to each article set can represent the path from the root node of the K-ary tree to each non-leaf node corresponding to each article set. Each leaf node in the K-ary tree corresponds to a candidate item.

According to the embodiment of the invention, the tree structure of the K-ary tree and the path sequence of the same candidate item on each binary tree can be illustrated by two K-ary trees and each K-ary tree is a balanced binary tree. For example, fig. 3(a) schematically shows a schematic diagram of a tree structure and a path sequence of a binary tree according to an embodiment of the present invention; fig. 3(b) schematically shows a tree structure and a path sequence of a binary tree according to another embodiment of the present invention.

Taking fig. 3(a) as an example, fig. 3(a) in the tree structure 300 may be the first to fourth levels from the top to the bottom, and nodes 300-1a to 300-15a in the respective levels. The first tier includes root node 300-1a, the second tier includes nodes 300-2a and 300-3a, the third tier includes nodes 300-4a through 300-7a, and the fourth tier includes nodes 300-8a through 300-15 a. In FIG. 3(a), the first level node 300-1a is the root node and corresponds to the candidate item set; the nodes 300-2a to 300-3a in the second layer are non-leaf nodes, and after balanced cluster analysis is performed on the articles in the candidate article set corresponding to 300-1a, the articles in the candidate article set corresponding to 300-1a are uniformly distributed to the nodes 300-2a and 300-3 a; the third-layer nodes 300-4a to 300-5a are non-leaf nodes, and after clustering analysis is performed on the articles in the article set corresponding to 300-2a, the articles in the article set corresponding to 300-2a are uniformly distributed to the nodes 300-4a and 300-5 a; similarly, the items in the item set corresponding to 300-3a are uniformly distributed to the nodes 300-6a and 300-7 a; the items in the item sets corresponding to the non-leaf nodes 300-4a to 300-7a are evenly distributed to the next level nodes 300-8a to 300-15a, respectively, of the non-leaf nodes 300-4a to 300-7 a. Nodes 300-8a through 300-15a are leaf nodes, respectively corresponding to 8 candidate items in the candidate item setiThat is to say that,i1,2, 3, …, 8.

Referring again to fig. 3(a), the direction representation of each node in fig. 3(a) is set in a preset direction from left to right of each node in fig. 3(a), the direction of the left direction is represented as "0", and the direction of the right direction is represented as "1". In FIG. 3(a), in addition to leaf nodes 300-8a to 300-15a, each of nodes 300-1a to 300-7a has two child nodes, i.e., two branches, which respectively represent "0" and "1" for two different directions. For example, the root node 300-1a has two branches 300-2a and 300-3a, and the direction of the directions 300-1a to 300-2a may be characterized as "0" and the direction of the directions 300-1a to 300-3a may be characterized as "1". Further, the root node 300-1a is the initial node in fig. 3(a), the root node 300-1a may be represented as "start", and "start" may be used as a special symbol for initializing the path sequence generation, and the path sequence from the root node 300-1a to the non-leaf node 300-2a may be represented as [ start, 0], and the path sequence from the root node 300-1a to the non-leaf node 300-3a may be represented as [ start, 1], so that the path sequence corresponding to the non-leaf node 300-2a is [ start, 0], and the path sequence corresponding to the non-leaf node 300-3a is [ start,1 ].

It should be noted that the tree structure in fig. 3(b) and the directional representations of the two branches of each node in different directions are the same as those in fig. 3 (a). In FIG. 3(b), nodes 300-8b through 300-15b are leaf nodes, and likewise correspond to 8 candidate items in the candidate item set, respectively, and 8 candidate items in the candidate item set in FIG. 3(a)iThe order of representation is different at the leaf nodes, i.e.,i5, 7, 1, 8, 2, 3, 4, 6.

According to the embodiment of the present invention, for the multi-branch tree with K being greater than 2, each node except for the leaf nodes has K direction representations, and the representation of the path sequence from the root node to each layer of nodes is similar to that in the above-mentioned binary tree, and is not described herein again.

According to the embodiment of the invention, a feature vector which is in accordance with personal preference of a user can be determined based on user feature information, and a path sequence corresponding to an article set of each non-leaf node is determined by combining the direction representation of node branches of each node except for leaf nodes in each K-tree in a forest formed by M K-trees, so that a target path from a root node to a leaf node is determined.

According to an embodiment of the present invention, the target path is a path of the target candidate item determined for each K-ary tree. After the target path is determined, the candidate item corresponding to the leaf node on the target path may be determined as the target candidate item. The target candidate item is an item to be recommended recommending the item to the target user. For example, referring again to fig. 3(a) -3 (b), assume that candidate item 4 is eventually the target candidate item. If the target path determined in fig. 3(a) is from the root node 300-1a to the leaf node 300-11a, the target path may be represented as [ start,0,1,1], and then the item 4 corresponding to the leaf node 300-11a is determined as a target candidate item; the target path determined in fig. 3(b) is from the root node 300-1b to the leaf node 300-14b, and the target path may be represented as [ start,1,1,0], and then the item 4 corresponding to the leaf node 300-14b is determined as the target candidate item. It should be noted that the root node "start" generates an initialized special symbol for the path sequence, and may not be displayed when the target path is finally output, so that the target path is finally displayed as [0,1,1] in fig. 3(a), and the target path is displayed as [1,1,0] in fig. 3 (b).

According to the embodiment of the invention, at least one target candidate item can be determined according to the user characteristic information and the path sequence corresponding to the item set for each K-ary tree. And determining the target candidate item determined on each K-ary tree as a target item to be recommended, determining a target recommended item recommended to the user from the target items to be recommended, and sending the target recommended item to the target user client so as to display the recommended item to the target user.

According to the embodiment of the invention, a forest is formed by M K-ary trees, and the generalization and diversity of the K-ary trees can be realized by carrying out candidate article based on different path sequences corresponding to the non-leaf nodes of each K-ary tree in the M K-ary trees arranged in the forest.

According to the embodiment of the invention, M K-ary trees are constructed, a target path from a root node to a leaf node is determined based on a path sequence and user characteristic information of an article set corresponding to each non-leaf node of the M K-ary trees, and a candidate article corresponding to the leaf node on the target path is determined as a target recommended article. The candidate item set is subjected to clustering analysis through a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees, so that the diversity and randomness of the M K-ary trees are guaranteed, the path sequence of the target candidate item is determined by the path sequence index of the item set of non-leaf nodes, and the information of the node at the upper layer of the K-ary tree structure is considered in the process of determining the target candidate item, so that the item at the division edge can be recommended to a user, and the item recommendation accuracy is improved.

Fig. 4 schematically shows a flowchart of a method for determining a candidate item corresponding to a leaf node as a target candidate item according to an embodiment of the present invention.

As shown in FIG. 4, the method 400 may include operations S410-S440.

In operation 410, feature extraction is performed on the user feature information to obtain a user feature vector.

In operation S420, based on the sequence from the root node to the leaf node, according to the user feature vector and the path sequence corresponding to the object set, the direction representation of the next-layer node of the current-layer node is predicted from the path sequence corresponding to the object set of the current-layer node, and a path sequence corresponding to the object set of the next-layer node is generated.

In operation S430, when the next-layer node is a non-leaf node, the next-layer node is used as a new current-layer node, direction representation of the next-layer node of the new current-layer node is repeatedly predicted by using the path sequence corresponding to the item set of the new current-layer node, and a path sequence corresponding to the item set of the next-layer node of the new current-layer node is generated.

In operation S440, in a case that the next-layer node is a leaf node, a target path from the root node to the leaf node is generated, and a candidate item corresponding to the leaf node on the target path is taken as a target candidate item.

According to the embodiment of the invention, the user characteristic information is input into an Encoder (Encoder) for processing, and the user characteristic information is subjected to characteristic extraction to obtain a user characteristic vector. The user feature vector may characterize a feature vector that conforms to the personal preferences of the user.

According to an embodiment of the present invention, a user feature vector is input to a decoder (decoder), and a target candidate item matching a user of the user feature vector is determined based on an order from a root node to a leaf node. Specifically, the root node is used as a current layer, the direction representation of the next layer node of the root node is predicted by combining K direction representations of the layer where the root node is located, and a path sequence corresponding to the article set of the next layer node is generated, wherein the path sequence corresponding to the article set of the next layer node is a path from the root node to the node. And then determining whether the next layer node is a non-leaf node or a leaf node. According to the embodiment of the present invention, when the next-layer node is a non-leaf node, the next-layer node is used as a new current layer, and the direction representation of the next-layer node of the new current-layer node is predicted based on the K direction representations of the new current-layer node according to the path sequence corresponding to the item set of the new current-layer node, so as to generate the path sequence corresponding to the item set of the next-layer node. The path sequence may be a path from the root node to a next-layer node of the new current-layer node, and the above process is repeatedly executed until the new next-layer node is a leaf node, and the candidate item corresponding to the leaf node is determined as a target candidate item matched with the user of the user feature vector.

According to the embodiment of the invention, when the next layer node is a leaf node, a path sequence from the root node to the leaf node is generated and determined as a target path, and the candidate article corresponding to the leaf node on the target path is determined as the target candidate article matched with the user of the user feature vector.

According to the embodiment of the present invention, the encoder may be any neural network structure for extracting features of the user feature information, and in the present invention, the structure of the encoder is not particularly limited.

According to an embodiment of the present invention, the Decoder may be a transform Decoder structure.

According to an embodiment of the present invention, determining a target recommended item according to N target candidate items may include: taking a union set of the N target candidate articles to generate a target candidate article set; determining a recommendation probability value of each target candidate item in the target candidate item set; based on a preset sorting rule, sorting the recommendation probability value of each target candidate item to obtain a sorting result of the recommendation probability values of the target candidate items; and determining the target recommended item according to the sequencing result of the recommendation probability values of the target candidate items.

According to the embodiment of the invention, for each K-ary tree, a target path from a root node to a leaf node in each K-ary tree is determined based on user characteristic information and a path sequence corresponding to an item set, and at least one target candidate item which is matched with a user of the user characteristic vector and corresponds to each K-ary tree is determined. And taking a union set of at least one target candidate item corresponding to each K-ary tree to generate a target candidate item set. The target candidate item set may include a plurality of target candidate items.

According to the embodiment of the invention, for each target candidate item in the target candidate item set, the recommendation probability of the target candidate item matched with the user of the user feature vector is calculated on the K-ary tree where each target candidate item is located, so as to obtain the recommendation probability value of each target candidate item in the target candidate item set on the corresponding K-ary tree.

According to the embodiment of the invention, the recommendation probability values of the same target candidate item on different K-ary trees are summed aiming at the same target candidate item on different K-ary trees, and the final recommendation probability value of the target candidate item is determined. For example, taking two K-ary trees as an example, the target candidate items matched with the user of the user feature vector are obtained on the first K-ary tree as a and B, and the target candidate items matched with the user of the user feature vector are obtained on the second K-ary tree as a and C. The target candidate item A, B, C is then merged to determine a target candidate item set { A, B, C }. And calculating the recommendation probability value of each target candidate item in the target candidate item set on the corresponding K-ary tree. A recommendation probability value for target candidate item B on the corresponding first K-ary tree may be calculated, and a recommendation probability value for target candidate item C on the corresponding second K-ary tree may be calculated. And respectively calculating recommendation probability values on a first K-ary tree and a second K-ary tree aiming at the target candidate item A, summing the recommendation probability values on the first K-ary tree and the second K-ary tree to obtain a sum value, and taking the sum value as the final recommendation probability value of the target candidate item A.

According to the embodiment of the present invention, the recommendation probability value of the target candidate item in the target candidate item set on each K-ary tree is calculated, which can be obtained by the following formula (1).

（1）

Wherein the content of the first and second substances,

in order for the user to be aware of the fact,

to and from the user

The target candidate item that is matched is,

for the user

The corresponding feature vector of the user is used,Hthe number of layers of each K-ary tree,hthe number of the first arbitrary layer in each K-ary tree,

is as followshRepresenting the direction corresponding to the layer node;

is as followshThe path sequence corresponding to the item set of the node above the layer,startto generate the initialized special symbols for the path sequence,catindicating a connection.

According to an embodiment of the present invention, the preset ordering rule may be a rule that orders from large to small. And sequencing the probability values of the target candidate articles obtained by calculation according to the descending order to obtain a sequencing result of the recommendation probability values of the target candidate articles, and recommending the target candidate articles corresponding to the preset first recommendation probability values in the sequencing result to the user as target recommendation articles. It should be noted that the preset first several recommended probability values may be set according to actual requirements, and are not limited herein.

According to the embodiment of the invention, whether the recommendation probability value of the target candidate object obtained by calculation meets the preset recommendation probability threshold value or not can be determined, and the target candidate object with the recommendation probability value larger than the preset recommendation probability value in the target candidate object set is recommended to the user as the target recommended object.

According to the embodiment of the invention, based on a hierarchical balanced clustering algorithm with randomness, the balanced clustering analysis is carried out on the candidate item set to construct M K-ary trees, and the method comprises the following steps: for constructing each K-ary tree, it can be constructed as follows. Specifically, feature extraction is carried out on candidate articles in the candidate article set to obtain a plurality of candidate article characterization vectors, wherein the candidate article characterization vectors are used as root nodes of a K-ary tree; determining K clustering centers of the candidate item characterization vectors; performing first balance clustering analysis on the multiple candidate item characterization vectors of the root node based on the K clustering centers and the multiple candidate item characterization vectors to obtain K clusters of the root node, wherein each cluster comprises the multiple candidate item characterization vectors with the same quantity; obtaining K clustering results of the root nodes according to the K clusters of the root nodes; taking the K clustering results of the root node as K non-leaf nodes of a first layer of the root node; and performing second balanced clustering analysis on each non-leaf node until each obtained clustering result comprises a candidate article characterization vector, and taking each clustering result comprising the candidate article characterization vector as a leaf node to obtain the K-ary tree.

According to the embodiment of the invention, the candidate item set is input into the trained scoring model, and the feature extraction is carried out on each candidate item in the candidate item set to obtain the characterization vector of each candidate item.

According to an embodiment of the present invention, the characterization vector of each candidate item may characterize specific information of each candidate item, for example, may include an item type, an item name, an item click rate, and the like. The scoring model may be a deep interest network model.

According to the embodiment of the invention, if the candidate item in the candidate item set cannot uniformly distribute the candidate item to each cluster center, the candidate item is randomly selected from the candidate item set and supplemented to the candidate item set, so that the candidate item in the candidate item set can be uniformly distributed to each cluster center. Wherein, when the candidate item is randomly reselected from the candidate item set to be supplemented into the candidate item set, the candidate item set can comprise repeated candidate items.

According to the embodiment of the invention, the K-ary tree can be constructed based on a balanced K-means clustering algorithm. Specifically, balanced cluster analysis is performed on root nodes for constructing the K-ary tree, namely, the characterization vectors of a plurality of candidate items are used as the root nodes for constructing the K-ary tree, K cluster centers are determined, one cluster center is randomly selected, the distance from each candidate item in the plurality of candidate items to the cluster center is calculated, the candidate item closest to the cluster center is distributed to the cluster center, and if L candidate items exist in the candidate item set, the candidate item closest to the cluster center is distributed to the cluster center

Each item is assigned to the cluster center. Based on the above mode, the rest

And evenly distributing the candidate articles to other K-1 clustering centers according to the mode to form K initial clusters. Aiming at each initial cluster, updating a new clustering center in each cluster, calculating the distance from a candidate article to a selected clustering center, distributing the candidate article closer to the new clustering center to obtain K clusters, iteratively repeating the process until the clustering center in each cluster is not changed any more to obtain final K clusters, finishing clustering the root nodes to obtain the clustering corresponding to the K clustersAnd (4) classifying the result. And taking the clustering results corresponding to the K clusters as K non-leaf nodes of the next layer of the root node. And based on the clustering mode, performing balanced clustering analysis on the candidate articles in the next layer of K non-leaf nodes until the layer where each clustering result only contains one candidate article is used as a leaf node of the K-ary tree, and thus, completing the construction of the balanced K-ary tree from the root node to the leaf node.

Construction of a balanced binary tree is exemplified, according to embodiments of the present invention, for example, in connection with fig. 3 (a). Assuming that the candidate article set can include 8 articles, taking the candidate article set as a first-layer root node 300-1a, based on the above balanced K-means clustering algorithm, where K takes a value of 2, determining 2 clustering centers of the candidate article set, randomly determining one clustering center, calculating a distance from each candidate article in the candidate article set to the clustering center, assigning the first 4 candidate articles closer to the clustering center, then calculating a distance from the remaining candidate articles to another clustering center, determining that the remaining 4 candidate articles are assigned to another clustering center, obtaining two clusters of the root node, iteratively updating the clustering centers of the two clusters until the clustering centers do not change any more, completing clustering of the root node, obtaining two final clusters as two clustering results of the root node, and taking the two clustering results of the root node as two non-leaf nodes of a second layer, i.e., 300-2a and 300-3a, non-leaf nodes 300-2a and 300-3a correspond to 4 candidate items, respectively. Based on the clustering mode, performing balanced clustering analysis on 4 candidate articles corresponding to each non-leaf node of the second layer to obtain 4 clusters, wherein each cluster contains 2 candidate articles, completing clustering on the nodes of the second layer, the obtained 4 clusters are clustering results of the nodes of the second layer, the clustering results of the nodes of the second layer are taken as four non-leaf nodes of the third layer, namely, the clustering results of the non-leaf nodes 300-2a are taken as two non-leaf nodes of the third layer as 300-4a and 300-5a, the clustering results of the non-leaf nodes 300-3a are taken as two non-leaf nodes of the third layer as 300-6a and 300-7a, and the 4 non-leaf nodes of the third layer are taken as 300-4a and 300-5a, 300-6a and 300-7 a; and performing balanced clustering analysis on 2 candidate objects corresponding to each non-leaf node of the third layer to obtain 8 clusters, wherein each cluster contains 1 candidate object, finishing clustering on the non-leaf nodes of the third layer, and taking the obtained 8 clusters as a clustering result of a node of the last layer, wherein each node of the last layer contains one candidate object, and taking the clustering result of the node of the last layer as a leaf node of a binary tree, namely 300-8a to 300-15a, so as to finish construction of the binary tree. In a binary tree, the branches of each node except the leaf node represent one directional representation, and 2 branches may represent 2 different directional representations.

According to the embodiment of the invention, the construction of the binary tree is only exemplary, and the construction of other K-ary trees can be obtained based on the clustering method.

FIG. 5 schematically shows a flowchart of an item recommendation model training method according to an embodiment of the invention.

As shown in FIG. 5, the method 500 may include operations S510-S520.

In operation S510, a training sample is obtained, where the training sample includes sample user characteristic information and a historical path sequence of a labeled sample item.

In operation S520, the item recommendation model to be trained is trained based on the training samples to obtain a trained item recommendation model, where a historical path sequence of the label sample item is obtained through a path sequence of a candidate item corresponding to the label sample item in M constructed K-ary trees, the M constructed K-ary trees are obtained by performing balanced clustering analysis on a candidate item set based on a hierarchical balanced clustering algorithm with randomness, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2.

According to an embodiment of the present invention, the training samples may be samples used to train a recommendation model for an item to be trained. The training samples may include sample user characteristic information and a historical path sequence of corresponding labeled sample items.

According to an embodiment of the present invention, the sample user characteristic information may include information of a part of items in the sample user purchased item set, and the tag sample item may be another part of items in the user purchased item set. For example, the item set purchased by the user includes 100 items, wherein information of 99 items is extracted as user characteristic information, and 1 item is extracted as a tag item.

According to the embodiment of the invention, the construction of each K-ary tree in the M K-ary trees is obtained by performing balanced clustering analysis on the candidate item set based on a hierarchical balanced clustering algorithm with randomness, and the specific steps can refer to relevant parts of an item recommendation method and are not repeated herein.

According to the embodiment of the invention, the training sample is input into the object recommendation model to be trained, the object recommendation model to be trained is trained based on the target function until the target function meets the preset ending condition, and the training is ended to obtain the trained object recommendation model.

According to an embodiment of the present invention, the item recommendation model may include an encoder and a Decoder, and the Decoder may be of a transform Decoder structure.

According to the embodiment of the present invention, the encoder may be any neural network structure for performing feature extraction on the sample user feature information, and in the present invention, the structure of the encoder is not particularly limited.

According to the embodiment of the invention, the encoder can be used for extracting the characteristics of the sample user characteristic information to obtain the sample user characteristic vector. The decoder may be configured to predict a path sequence for a sample tagged item that matches a sample user of sample user characteristic information.

According to the embodiment of the invention, training the recommendation model of the article to be trained based on the training sample comprises the following steps: performing feature extraction on sample user feature information in a training sample aiming at each K-ary tree to obtain a sample user feature vector; determining an objective function value according to the sample user characteristic vector and the historical path sequence of the label sample article; and adjusting the model parameters of the recommendation model of the object to be trained based on the objective function value.

According to the embodiment of the invention, the characteristic extraction is carried out on the sample user characteristic information based on the encoder in the recommendation model of the object to be trained, so as to obtain the sample user characteristic vector.

According to the embodiment of the invention, the objective function value can be calculated by an objective function, and the objective function is shown as formula (2).

（2）

Wherein the content of the first and second substances,

in order for the user of the sample to be,

to and from the sample user

The corresponding sample article of the label is,

for sample users

The corresponding sample user feature vector is then used,Hthe number of layers of each K-ary tree,hthe number of the first layer in each K-ary tree,

is a firsthRepresenting the direction corresponding to the layer node;

is as followshThe path sequence corresponding to the node on the layer,startto generate the initialized special symbols for the path sequence,catindicating a connection.

According to the embodiment of the invention, the historical path sequence of the label sample object and the sample user characteristic information are input into a decoder of the recommendation model of the object to be trained, direction representation training is carried out on nodes of each layer of the K-ary tree based on an objective function, an objective function value aiming at the recommendation model of the object to be trained is obtained, and model parameters of the recommendation model of the object to be trained are adjusted according to the function value.

According to an embodiment of the present invention, the historic path sequence of the label sample item may include a given historic path sequence of each layer of the label sample item in the K-ary tree.

According to the embodiment of the invention, the determining the objective function value according to the sample user feature vector and the historical path sequence of the label sample item comprises the following steps: predicting the probability value of the path direction representation of the next layer of each layer in the K-ary tree corresponding to the sample user feature vector according to the sample user feature vector and the given historical path sequence of each layer in the K-ary tree to obtain the probability value of the path direction representation of each layer in the K-ary tree; and determining an objective function value according to the probability value represented by the path direction of each layer.

According to an embodiment of the present invention, for each K-ary tree, for example, reference may be made to four levels (of) of fig. 3(a)H= 4) binary tree as an example. Hypothetical label sample item

Corresponding to candidate items in the binary tree

A path sequence of 4, i.e., [0, 1]]Then the path sequence [0, 1] can be passed]The given historical path sequence of the tag sample item at each level in the binary tree is known.

According to an embodiment of the invention, a sample user feature vector and a historical path sequence [0, 1] of a labeled sample item are combined]And inputting the data to a decoder of the recommendation model of the object to be trained, regarding each layer of the binary tree, taking the root node of the binary tree as an initial node, and when h =0, the data is the first layer of the binary tree, namely the root node of the binary tree, and because the root node is the initial node, the probability value for predicting the root node does not exist. When h =1, which is the second level of the binary tree, according to the history path sequence of the node at the upper level of the second level of the binary tree, that is, the history path sequence of the node at the first level (root node), since the root node is the initial node of the binary tree, there is no history path sequence since the root node is the initial node of the binary tree

Corresponding to the historical path sequence, the root node may be represented by "start" to generate the initialized special symbol as the path sequence. Predicting a probability value of a direction representation of a second level node, i.e. a path direction representation from a root node to the second level node, with a root node "start", based on an objective function

The maximum probability value of "0" or "1". Based on the objective function, the historical path sequence corresponding to the second layer node is utilized in a similar way

Is [ start, 0]]Time prediction h =2 level, i.e. path direction characterization corresponding to a node of the third level of the binary tree

A maximum probability value of "0" or "1"; path history sequence corresponding to nodes in third layer

Is [ start,0, 1]]Temporal prediction h =3 levels, i.e. path direction characterization corresponding to the fourth level node (leaf node) of the binary tree

The maximum probability value of "0" or "1". And summing the maximum probability values of the path direction representations obtained by calculation of each layer, and determining the objective function value.

According to the embodiment of the present invention, it should be noted that, in the foregoing, a manner of predicting the probability value of the path direction representation of the next layer of each layer according to the given historical path sequence of each layer in the binary tree is exemplary, and a manner of predicting the path direction representation of the next layer of each layer of other K-ary trees is similar to that of the binary tree, and details are not repeated here.

According to the embodiment of the invention, an item recommendation model of a forest is formed by M K-ary trees, and the direction representations of each layer of the K-ary trees are consistent, namely, the K-ary trees have K direction representations in different directions. The path sequence of the next layer is represented and predicted in the same direction of each layer, the parameter quantity and the memory consumption of the model are reduced, and the training time cost and the occupied storage space of the model are reduced.

Fig. 6 is a block diagram schematically showing the configuration of an item recommendation apparatus according to an embodiment of the present invention.

As shown in fig. 6, the apparatus 600 may include: a first obtaining module 610, a constructing module 620, a first determining module 630, and a second determining module 640.

A first obtaining module 610, configured to obtain the candidate item set and the user characteristic information of the target user.

A constructing module 620, configured to perform balanced clustering analysis on the candidate item sets based on a hierarchical balanced clustering algorithm with randomness, and construct M K-ary trees, where each non-leaf node of each K-ary tree corresponds to one item set, each item set corresponds to one path sequence, the path sequence corresponding to each item set characterizes a path from a root node of the K-ary tree to each non-leaf node corresponding to each item set, each leaf node of each K-ary tree corresponds to one candidate item, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2.

A first determining module 630, configured to determine, for each K-ary tree, a target path from the root node to a leaf node according to the user feature information and a path sequence corresponding to the item set, and use a candidate item corresponding to the leaf node on the target path as a target candidate item.

A second determining module 640, configured to determine a target recommended item according to N target candidate items, where N is a positive integer greater than or equal to M.

According to an embodiment of the present invention, the first determining module 630 may include: the device comprises a first extraction submodule, a first prediction submodule, a second prediction submodule and a first determination submodule.

And the first extraction submodule is used for extracting the characteristics of the user characteristic information to obtain a user characteristic vector.

And the first prediction sub-module is used for predicting the direction representation of the next layer node of the current layer node according to the user characteristic vector and the path sequence corresponding to the object set of the current layer node based on the sequence from the root node to the leaf node, and generating the path sequence corresponding to the object set of the next layer node.

And the second prediction submodule is used for taking the next-layer node as a new current-layer node under the condition that the next-layer node is a non-leaf node, repeatedly executing the direction representation of the next-layer node of the new current-layer node predicted by the path sequence corresponding to the item set of the new current-layer node, and generating the path sequence corresponding to the item set of the next-layer node of the new current-layer node.

And the first determining submodule is used for generating a target path from the root node to the leaf node under the condition that the next layer node is the leaf node, and taking the candidate article corresponding to the leaf node on the target path as the target candidate article.

According to an embodiment of the present invention, the second determining module 640 may include: the device comprises a generating submodule, a second determining submodule, a sequencing submodule and a third determining submodule.

And the generation submodule is used for taking a union set of the N target candidate items and generating a target candidate item set.

And the second determining submodule is used for determining the recommendation probability value of each target candidate item in the target candidate item set.

And the sorting submodule is used for sorting the recommendation probability value of each target candidate item based on a preset sorting rule to obtain a sorting result of the recommendation probability value of the target candidate item.

And the third determining submodule is used for determining the target recommended item according to the sequencing result of the recommendation probability values of the target candidate items.

According to an embodiment of the invention, the building module 620 may include: the second extraction submodule, the fourth determination submodule, the clustering submodule, the obtaining submodule, the fifth determination submodule and the sixth determination submodule.

And the second extraction submodule is used for performing feature extraction on the candidate articles in the candidate article set to obtain a plurality of candidate article characterization vectors, wherein the plurality of candidate article characterization vectors are used as root nodes of the K-ary tree.

And the fourth determining submodule is used for determining K clustering centers of the candidate item characterization vectors.

And the clustering submodule is used for carrying out first balanced clustering analysis on the candidate item characterization vectors of the root node based on the K clustering centers and the candidate item characterization vectors to obtain K clusters of the root node, wherein each cluster comprises a plurality of candidate item characterization vectors with the same quantity.

And the obtaining submodule is used for obtaining K clustering results of the root node according to the K clusters of the root node.

And the fifth determining submodule is used for taking the K clustering results of the root node as K non-leaf nodes of the first layer of the root node.

And the sixth determining submodule is used for carrying out second balanced clustering analysis on the K non-leaf nodes until each obtained clustering result comprises a candidate article characterization vector, and taking each clustering result comprising one candidate article characterization vector as a leaf node to obtain the K-ary tree.

Fig. 7 schematically shows a block diagram of the structure of an item recommendation model training apparatus according to an embodiment of the present invention.

As shown in fig. 7, the apparatus 700 may include: a second acquisition module 710 and a training module 720.

A second obtaining module 710, configured to obtain a training sample, where the training sample includes sample user feature information and a historical path sequence of a labeled sample item.

The training module 720 is configured to train the item recommendation model to be trained based on the training samples to obtain a trained item recommendation model, where a historical path sequence of the label sample item is obtained through a path sequence of a candidate item corresponding to the label sample item in M K-ary trees that are obtained by performing balanced clustering analysis on a candidate item set based on a hierarchical balanced clustering algorithm with randomness. The item recommendation model comprises an encoder and a Decoder, and the Decoder is of a transform Decoder structure.

According to an embodiment of the invention, training module 720 may include: a third extraction sub-module, a seventh determination sub-module, and an adjustment sub-module.

And the third extraction submodule is used for extracting the characteristics of the sample user characteristic information in the training sample aiming at each K-ary tree to obtain a sample user characteristic vector.

And the seventh determining sub-module is used for determining the objective function value according to the sample user feature vector and the historical path sequence of the label sample item, wherein the historical path sequence of the label sample item comprises a given path historical sequence of each layer of the label sample item in the K-ary tree.

And the adjusting submodule is used for adjusting the model parameters of the recommendation model of the object to be trained based on the objective function value.

According to an embodiment of the present invention, the seventh determining sub-module may include: a prediction unit and a determination unit.

And the prediction unit is used for predicting the probability value of the path direction representation of the next layer of each layer in the K-ary tree corresponding to the sample user feature vector according to the sample user feature vector and the given path history sequence of each layer in the K-ary tree to obtain the probability value of the path direction representation of each layer in the K-ary tree.

And the determining unit is used for determining the objective function value according to the probability value represented by the path direction of each layer.

According to the embodiment of the present invention, any plurality of the first obtaining module 610, the constructing module 620, the first determining module 630 and the second determining module 640, or the second obtaining module 710 and the training module 720 may be combined into one module to be implemented, or any one of them may be split into a plurality of modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the first obtaining module 610, the constructing module 620, the first determining module 630 and the second determining module 640, or the second obtaining module 710 and the training module 720 may be at least partially implemented as a hardware circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented by hardware or firmware in any other reasonable manner of integrating or packaging a circuit, or implemented by any one of three implementations of software, hardware and firmware, or an appropriate combination of any of them. Alternatively, at least one of the first obtaining module 610, the building module 620, the first and second determining

modules

630, 640 or the second obtaining module 710 and the training module 720 may be at least partially implemented as a computer program module, which when executed, may perform the corresponding functions.

As shown in fig. 8, an electronic device 800 according to an embodiment of the present invention includes a processor 801 which can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 802 or a program loaded from a storage section 808 into a Random Access Memory (RAM) 803. The processor 801 may include, for example, a general purpose microprocessor (e.g., CPU), an instruction set processor and/or related chip sets and/or a special purpose microprocessor (e.g., Application Specific Integrated Circuit (ASIC)), among others. The processor 801 may also include onboard memory for caching purposes. The processor 801 may comprise a single processing unit or a plurality of processing units for performing the different actions of the method flows according to embodiments of the present invention.

In the RAM 803, various programs and data necessary for the operation of the electronic apparatus 800 are stored. The processor 801, the ROM 802, and the RAM 803 are connected to each other by a bus 804. The processor 801 performs various operations of the method flow according to the embodiment of the present invention by executing programs in the ROM 802 and/or the RAM 803. Note that the program may also be stored in one or more memories other than the ROM 802 and the RAM 803. The processor 801 may also perform various operations of method flows according to embodiments of the present invention by executing programs stored in the one or more memories.

Electronic device 800 may also include input/output (I/O) interface 805, input/output (I/O) interface 805 also connected to bus 804, according to an embodiment of the invention. The electronic device 800 may also include one or more of the following components connected to the I/O interface 805: an input portion 806 including a keyboard, a mouse, and the like; an output section 807 including components such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage section 808 including a hard disk and the like; and a communication section 809 including a network interface card such as a LAN card, a modem, or the like. The communication section 809 performs communication processing via a network such as the internet. A drive 810 is also connected to the I/O interface 805 as necessary. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 810 as necessary, so that a computer program read out therefrom is mounted on the storage section 808 as necessary.

The present invention also provides a computer-readable storage medium, which may be embodied in the device/apparatus/system described in the above embodiments; or may exist alone without being assembled into the device/apparatus/system. The computer-readable storage medium carries one or more programs which, when executed, implement the method according to an embodiment of the present invention.

According to embodiments of the present invention, the computer readable storage medium may be a non-volatile computer readable storage medium, which may include, for example but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present invention, a computer 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. For example, according to embodiments of the present invention, a computer-readable storage medium may include the ROM 802 and/or the RAM 803 described above and/or one or more memories other than the ROM 802 and the RAM 803.

Embodiments of the invention also include a computer program product comprising a computer program comprising program code for performing the method illustrated in the flow chart. When the computer program product runs in a computer system, the program code is used for causing the computer system to realize the item recommendation method provided by the embodiment of the invention.

The computer program performs the above-described functions defined in the system/apparatus of the embodiment of the present invention when executed by the processor 801. The above described systems, devices, modules, units, etc. may be implemented by computer program modules according to embodiments of the invention.

In one embodiment, the computer program may be hosted on a tangible storage medium such as an optical storage device, a magnetic storage device, or the like. In another embodiment, the computer program may also be transmitted in the form of a signal on a network medium, distributed, downloaded and installed via communication section 809, and/or installed from removable media 811. The computer program containing program code may be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

In such an embodiment, the computer program can be downloaded and installed from a network through the communication section 809 and/or installed from the removable medium 811. The computer program, when executed by the processor 801, performs the above-described functions defined in the system of the embodiment of the present invention. The above described systems, devices, apparatuses, modules, units, etc. may be implemented by computer program modules according to embodiments of the present invention.

According to embodiments of the present invention, program code for executing a computer program provided by embodiments of the present invention may be written in any combination of one or more programming languages, and in particular, the computer program may be implemented using a high level procedural and/or object oriented programming language, and/or an assembly/machine language. The programming language includes, but is not limited to, programming languages such as Java, C + +, python, the "C" language, or the like. The program code may execute entirely on the user computing device, partly on the user device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, 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., through the internet using an internet service provider).

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

It will be appreciated by a person skilled in the art that features described in the various embodiments of the invention may be combined in various ways and/or combinations, even if such combinations or combinations are not explicitly described in the invention. In particular, various combinations and/or subcombinations of the features described in various embodiments of the invention may be made without departing from the spirit and teachings of the invention. All such combinations and/or associations fall within the scope of the present invention.

The embodiments of the present invention have been described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Although the embodiments are described separately above, this does not mean that the measures in the embodiments cannot be used in advantageous combination. The scope of the invention is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be devised by those skilled in the art without departing from the scope of the invention, and these alternatives and modifications are intended to fall within the scope of the invention.

Claims

1. An item recommendation method comprising:

acquiring user characteristic information of a candidate item set and a target user;

performing balanced clustering analysis on the candidate item sets based on a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees, wherein each non-leaf node of each K-ary tree corresponds to one item set, each item set corresponds to one path sequence, the path sequence corresponding to each item set represents a path from a root node of the K-ary tree to each non-leaf node corresponding to each item set, each leaf node of each K-ary tree corresponds to one candidate item, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2;

for each K-ary tree, determining a target path from the root node to the leaf node according to the user characteristic information and a path sequence corresponding to the item set, and taking a candidate item corresponding to the leaf node on the target path as a target candidate item;

and determining a target recommended item according to N target candidate items, wherein N is a positive integer greater than or equal to 1.

2. The method according to claim 1, wherein determining a target path from the root node to the leaf node according to the user feature information and a path sequence corresponding to the item set, and taking a candidate item corresponding to the leaf node on the target path as a target candidate item comprises:

extracting the features of the user feature information to obtain a user feature vector;

predicting the direction representation of a next layer node of the current layer node by the path sequence corresponding to the article set of the current layer node according to the user feature vector and the path sequence corresponding to the article set based on the sequence from the root node to the leaf node, and generating the path sequence corresponding to the article set of the next layer node;

under the condition that the next-layer node is a non-leaf node, taking the next-layer node as a new current-layer node, repeatedly executing the direction representation of predicting the new next-layer node of the current-layer node by using the path sequence corresponding to the new article set of the current-layer node, and generating a new path sequence corresponding to the article set of the next-layer node of the current-layer node;

and under the condition that the next layer node is a leaf node, generating a target path from a root node to the leaf node, and taking a candidate article corresponding to the leaf node on the target path as a target candidate article.

3. The method of claim 1, wherein said determining a target recommended item from the N target candidate items comprises:

taking a union set of the N target candidate articles to generate a target candidate article set;

determining a recommendation probability value for each of the target candidate items in the set of target candidate items;

based on a preset sorting rule, sorting the recommendation probability value of each target candidate item to obtain a sorting result of the recommendation probability values of the target candidate items;

and determining the target recommended item according to the sequencing result of the recommendation probability values of the target candidate items.

4. The method according to claim 1, wherein the performing balanced clustering analysis on the candidate item set based on a hierarchical balanced clustering algorithm with randomness to construct M K-ary trees comprises:

aiming at each K-branch tree, the method comprises the following steps of,

performing feature extraction on candidate articles in the candidate article set to obtain a plurality of candidate article characterization vectors, wherein the candidate article characterization vectors are used as root nodes of the K-ary tree;

determining K clustering centers of the candidate item characterization vectors;

performing first balanced clustering analysis on the candidate item characterization vectors of the root node based on the K clustering centers and the candidate item characterization vectors to obtain K clusters of the root node, wherein each cluster comprises a plurality of candidate item characterization vectors with the same quantity;

obtaining K clustering results of the root node according to the K clusters of the root node;

taking the K clustering results of the root node as K non-leaf nodes of a first layer of the root node;

and performing second balanced clustering analysis on each non-leaf node until each obtained clustering result comprises one candidate item characterization vector, and taking each clustering result comprising one candidate item characterization vector as a leaf node to obtain the K-ary tree.

5. An item recommendation model training method comprises the following steps:

obtaining a training sample, wherein the training sample comprises sample user characteristic information and a historical path sequence of a label sample article;

training an item recommendation model to be trained based on the training samples to obtain a trained item recommendation model, wherein the historical path sequence of the label sample item is obtained through the path sequence of a candidate item corresponding to the label sample item in M constructed K-ary trees, the M constructed K-ary trees are obtained by performing balanced clustering analysis on a candidate item set based on a hierarchical balanced clustering algorithm with randomness, K is an integer greater than or equal to 2, and M is an integer greater than or equal to 2.

6. The method of claim 5, wherein training an item recommendation model to be trained based on the training samples comprises:

performing feature extraction on the sample user feature information in the training sample aiming at each K-ary tree to obtain a sample user feature vector;

determining an objective function value according to the sample user feature vector and the historical path sequence of the label sample article;

and adjusting the model parameters of the recommendation model of the to-be-trained object based on the objective function value.

7. The method of claim 6, wherein the historical path sequence of the labeled sample item comprises a given historical path sequence for each level of the labeled sample item in the K-ary tree;

determining an objective function value according to the sample user feature vector and the historical path sequence of the labeled sample item, including:

predicting the probability value of the path direction representation of the next layer of each layer in the K-ary tree corresponding to the sample user feature vector according to the sample user feature vector and the given historical path sequence of each layer in the K-ary tree to obtain the probability value of the path direction representation of each layer in the K-ary tree;

and determining the objective function value according to the probability value represented by the path direction of each layer.

8. The method of claim 5, wherein the item recommendation model comprises an encoder and a decoder.

9. The method of claim 8, wherein the Decoder is a transform Decoder structure.

10. An electronic device, comprising:

one or more processors;

a memory for storing one or more programs,

wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-9.