CN111046285B

CN111046285B - Recommendation ordering determining method, device, server and storage medium

Info

Publication number: CN111046285B
Application number: CN201911263832.1A
Authority: CN
Inventors: 刘记平
Original assignee: Rajax Network Technology Co Ltd
Current assignee: Rajax Network Technology Co Ltd
Priority date: 2019-12-11
Filing date: 2019-12-11
Publication date: 2023-04-25
Anticipated expiration: 2039-12-11
Also published as: CN111046285A

Abstract

The embodiment of the invention provides a recommendation ordering determining method, a recommendation ordering determining device, a server and a storage medium, wherein the method comprises the following steps: acquiring a behavior log of a target user; analyzing a behavior log through a processor, and determining a behavior sequence of a target user, wherein the behavior sequence at least comprises an object sequence corresponding to an object of the target user for executing the set behavior, and the objects in the object sequence are arranged according to the time sequence of the target user for executing the set behavior; determining corresponding behavior characteristics of a target object to be recommended and a behavior sequence through a processor, wherein the behavior characteristics at least represent the matching degree of the target object and the behavior sequence; determining, by a processor, a recommended ranking value of the target object for the target user based at least on the behavioral characteristics; the recommendation ordering value is related to a recommendation ordering of the target object for the target user. The method and the device can determine the object recommendation sequence aiming at the individuation of the user, and provide a basis for improving the accuracy of object recommendation.

Description

Recommendation ordering determining method, device, server and storage medium

Technical Field

The embodiment of the invention relates to the technical field of data processing, in particular to a recommendation ordering determining method, a recommendation ordering determining device, a server and a storage medium.

Background

In the field of object recommendation, the online internet platform can recommend objects to a user by determining the recommendation sequence of the objects so as to recommend the sequences of the objects; at present, the recommendation ordering of the objects is determined mainly based on the overall user preference degree of the objects in the overall user layer, and the higher the overall user preference degree is, the earlier the recommendation ordering of the objects is.

The above manner determines the recommendation sequence of the object based on the overall preference degree of the user of the object, and the recommendation sequence of the object cannot be determined for the user in a personalized way, which results in the fact that the personalized object cannot be accurately recommended for the user, and the accuracy of the object recommendation is lower.

Disclosure of Invention

In view of the above, embodiments of the present invention provide a recommendation order determining method, apparatus, server and storage medium, so as to determine an object recommendation order for user individuation, and provide a basis for improving accuracy of object recommendation.

In order to achieve the above object, the embodiment of the present invention provides the following technical solutions:

in a first aspect, an embodiment of the present invention provides a recommendation ordering determining method, including:

Acquiring a behavior log of a target user;

analyzing the behavior log through a processor to determine a behavior sequence of the target user, wherein the behavior sequence at least comprises an object sequence corresponding to an object of the target user for executing the set behavior, and the objects in the object sequence are arranged according to the time sequence of the target user for executing the set behavior;

determining behavior characteristics of a target object to be recommended and corresponding to the behavior sequence through a processor, wherein the behavior characteristics at least represent the matching degree of the target object and the behavior sequence;

determining, by a processor, a recommended ranking value of the target object for the target user based at least on the behavioral characteristics; the recommendation ordering value is related to a recommendation ordering of the target object for the target user.

With reference to the first aspect, in a first implementation manner of the first aspect, the behavior sequence further includes an object type sequence; the object type sequence comprises an object type corresponding to an object of which the target user executes the set behavior.

With reference to the first aspect or the first implementation manner of the first aspect, in a second implementation manner of the first aspect, the determining, by the processor, a behavior feature of the target object to be recommended corresponding to the behavior sequence includes:

Determining, by the processor, an attention score of the target object for the sequence of behaviors, the attention score being the behavior feature.

With reference to the first aspect or the first implementation manner of the first aspect, in a third implementation manner of the first aspect, the determining, by the processor, a recommended ranking value of the target object for the target user according to at least the behavior feature includes:

at least discretizing the statistical characteristics of the target object by the processor to obtain discretized characteristics;

and determining, by the processor, a recommended ranking value of the target object for the target user according to the behavior feature and the discretization feature.

With reference to the third implementation manner of the first aspect, in a fourth implementation manner of the first aspect, the performing, by the processor, discretizing at least the statistical feature of the target object to obtain a discretized feature includes:

discretizing the statistical features of the target object with at least one feature to obtain discretized features; the at least one feature includes at least one of: the method comprises the steps of attribute characteristics of a target user, attribute characteristics of an object, interaction class characteristics of the target user and recommendation upper and lower class characteristics.

With reference to the third implementation manner of the first aspect, in a fifth implementation manner of the first aspect, the determining, by the processor, a recommended ranking value of the target object for the target user according to the behavior feature and the discretized feature includes:

extracting first-order and second-order feature information of the discretization features;

combining the discretization features with the behavior features, and extracting hidden abstract features of the combined features;

connecting the hidden layer abstract feature and the first-order and second-order feature information to obtain a connected feature;

and determining a recommended sorting value corresponding to the connected features.

With reference to the first implementation manner of the first aspect, in a sixth implementation manner of the first aspect, the analyzing, by the processor, the behavior log, and determining a behavior sequence of the target user includes:

determining objects of the target user, which are ordered according to the time sequence and execute the set behaviors, from an online dotting log of the target user to form the object sequence; and determining a sequence of object types corresponding to the object of the user executing the set behavior from the offline log of the target user to form the sequence of object types.

In a second aspect, an embodiment of the present invention provides a recommendation ordering determining apparatus, including:

the log acquisition module is used for acquiring the behavior log of the target user;

the sequence determining module is used for analyzing the behavior log through the processor to determine a behavior sequence of the target user, wherein the behavior sequence at least comprises an object sequence corresponding to an object of the target user for executing the set behavior, and the objects in the object sequence are arranged according to the time sequence of the target user for executing the set behavior;

the behavior characteristic determining module is used for determining behavior characteristics of a target object to be recommended and corresponding to the behavior sequence through the processor, wherein the behavior characteristics at least represent the matching degree of the target object and the behavior sequence;

the ranking determining module is used for determining a recommended ranking value of the target object for the target user at least according to the behavior characteristics through the processor; the recommendation ordering value is related to a recommendation ordering of the target object for the target user.

In a third aspect, an embodiment of the present invention provides a server, including at least one memory storing one or more computer-executable instructions and at least one processor invoking the one or more computer-executable instructions to perform the recommendation ordering determining method according to any of the above.

In a fourth aspect, embodiments of the present invention provide a storage medium storing one or more computer-executable instructions for performing the recommendation ordering determining method of any of the above.

In the recommendation ordering determining method provided by the embodiment of the invention, a server can acquire a behavior log of a target user, and the behavior log is analyzed by a processor to determine a behavior sequence of the target user, wherein the behavior sequence at least comprises an object sequence corresponding to an object of the target user for executing set behaviors, and the objects in the object sequence are arranged according to the time sequence of the target user for executing set behaviors; therefore, the embodiment of the invention can embody the object favored by the target in real time within a certain time through the object sequence in the behavior sequence, so that the behavior sequence at least shows the favored object of the target user in the online internet platform along with the time; when determining recommendation ordering for the target object to be recommended, the server can determine the behavior characteristics of the target object corresponding to the behavior sequence through the processor, wherein the behavior characteristics at least represent the matching degree of the target object and the behavior sequence, so that the recommendation ordering value of the target object for the target user is determined through the processor according to at least the behavior characteristics; because the recommendation ordering value is related to the recommendation ordering of the target object for the target user and the recommendation ordering value is determined based on the behavior sequence, based on the recommendation ordering value, the recommendation ordering of the determined target object for the target user can be matched with the preference degree of the target user, the recommendation ordering of the target user for the favorite object of the online interconnection platform can be guaranteed to be forward, and personalized determination of the recommendation ordering of the target object for the user is realized.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings may be obtained according to the provided drawings without inventive effort to a person skilled in the art.

FIG. 1 is a flowchart of a recommendation ordering determining method according to an embodiment of the present invention;

FIG. 2 is another flowchart of a recommendation ordering determining method according to an embodiment of the present invention;

FIG. 3 is a flowchart of determining a recommendation-ranking value provided by an embodiment of the present invention;

FIG. 4 is a schematic diagram of a model according to an embodiment of the present invention;

FIG. 5 is a schematic illustration of a recommendation ordering determination process based on the model shown in FIG. 4;

FIG. 6 is an exemplary diagram of an extraction behavior sequence provided by an embodiment of the present invention;

FIG. 7 is a block diagram of a recommendation order determining apparatus provided in an embodiment of the present invention;

FIG. 8 is another block diagram of a recommendation order determining apparatus according to an embodiment of the present invention;

fig. 9 is a hardware block diagram of a server according to an embodiment of the present invention.

Detailed Description

Whether the object is favored by the user can be determined by whether the user executes a setting action for the object, taking e-commerce, online take-away and other scenes as an example, the setting action can be such as the action that the user purchases the object, buys the object again, clicks the object, collects the object and the like to express the user's preference; the statistical characteristics of the object can be obtained by counting the situation that the object is executed by all users to set the behavior, and the overall user preference degree of the object on the overall user level can be reflected by the statistical characteristics of the object; for example, the statistical features of the object may be, for example, a monthly sales volume, a repurchase rate, a number of user clicks of a last certain day, etc., and based on the statistical features of the object, the overall user preference degree of the object on the overall user plane may be determined, so as to determine the recommendation ranking of the object according to the overall user preference degree of the object, so that the recommendation ranking of the object with a higher overall user preference degree is more advanced;

although the above-mentioned recommendation ordering determining method may order the highest object such as the monthly sales volume, the repurchase rate, the number of clicks of the user, etc., the statistical feature is actually that the historical behaviors of the whole user for the object are described in a summation, averaging, etc., for example, the monthly sales volume is that the monthly purchasing behaviors of the whole user for the object are described in a summation, it is seen that the object ordered by the above-mentioned recommendation ordering determining method is only the object with higher overall preference of the user; however, for a single user, the objects determined to be ranked ahead in the above manner are not necessarily objects with higher preference of the single user, so that the ranking cannot be recommended for the determined objects personalized for the user in the above manner, which results in that the objects personalized for the user cannot be accurately recommended, and the accuracy of the object recommendation is lower.

Based on the above, the embodiment of the invention provides an improved recommendation ordering determination scheme to determine the object recommendation ordering aiming at user individuation, and provides a basis for improving the accuracy of object recommendation. The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

As an optional implementation, fig. 1 shows a flowchart of a recommendation order determining method provided by an embodiment of the present invention, where the method shown in fig. 1 may alternatively be applied to an online internet platform, for example, the method shown in fig. 1 may be performed by a server of the online internet platform, and in an optional implementation, the method shown in fig. 1 may be performed by a server corresponding to a recommendation system of the online internet platform; referring to fig. 1, the process may include:

step S100, obtaining a behavior log of a target user.

The target user can be any user of an online internet platform, such as any user of an online take-out platform, and when recommending objects for the target user, the embodiment of the invention aims to provide personalized object recommendation ordering for the target user, so as to provide a basis for improving the accuracy of object recommendation.

The behavior log of the target user may record behavior information of the target user on the online internet platform, for example, a behavior type of the target user on the online internet platform, an object targeted by the behavior, and the like.

Step S110, analyzing the behavior log through a processor to determine a behavior sequence of the target user, wherein the behavior sequence at least comprises an object sequence corresponding to an object of the target user for executing the set behavior, and the objects in the object sequence are arranged according to the time sequence of the target user for executing the set behavior.

Based on the behavior information of the target user on the online internet platform recorded by the behavior log, the embodiment of the invention can analyze the behavior log through the processor to obtain a behavior sequence of the target user, and the behavior sequence reflects the object favored by the user on the online internet platform;

optionally, the behavior sequence of the target user may at least indicate the objects favored by the user on the online interconnection platform according to the time sequence, so as to reflect the objects favored by the target user in real time within a certain time; the object is represented by the user preference by executing the set behavior on the object by the user, the behavior sequence of the target user at least comprises an object sequence corresponding to the object of the set behavior executed by the target user, and the objects in the object sequence are arranged according to the time sequence of the set behavior executed by the target user;

In an optional implementation, according to the behavior log of the target user, the embodiment of the invention can sequentially record the objects of the set behaviors executed by the target user according to the time sequence of the behaviors executed by the target user, thereby forming the object sequence;

optionally, taking e-commerce, online take a take for example, the setting behavior may be a purchasing behavior, a re-purchasing (repeated purchasing) behavior, a clicking behavior, a collecting behavior, etc. which represent the preference of the user, and the specific form of the setting behavior may be set according to the actual situation, and the embodiment of the present invention is not limited; in one example, according to the behavior log of the target user, the embodiment of the invention can sequentially record the objects purchased, purchased again, clicked, and/or collected by the target user according to the time sequence of the behavior executed by the target user, so as to form an object sequence; in an alternative example, the object may be a merchant and/or commodity of an online internet platform in the context of e-commerce, online take-away, etc.

Optionally, the object sequence may be part of the content of the behavior sequence, and the embodiment of the present invention may further determine a sequence of object types corresponding to the object where the target user performs the set behavior on the online internet platform (referred to as an object type sequence in the embodiment of the present invention), where the object type sequence may include an object type corresponding to the object where the target user performs the set behavior on the online internet platform, so that the object type corresponding to the object favored by the target user is reflected by the object type sequence; it can be understood that the object sequence may embody an object that is favored by the target user in real time within a certain time of the online internet platform, however, the object sequence cannot cover a long-tail object of the online internet platform, and the long-tail object refers to an online internet platform such as a newly-online and cold-door object, and in order to mine an object that is favored by the target user and is possibly contained in the long-tail object, the embodiment of the invention can mine an object that is favored by the target user and is contained in the long-tail object through an object type corresponding to the object that is favored by the target user in the object type sequence;

It can be understood that the behavior sequence represents the object favored by the target user on the online internet platform, and the embodiment of the invention reflects the object favored by the target user on the online internet platform based on the object type (represented by the object type sequence) corresponding to the object favored by the target user besides the object favored by the target user on the online internet platform in real time (represented by the object sequence); therefore, in an optional implementation, the behavior sequence in the embodiment of the invention can include the object sequence and the object type sequence, so that the behavior sequence can more comprehensively and accurately reflect the objects favored by the user on the online internet platform; of course, the use of the sequence of object types is only an optional implementation of an embodiment of the invention, which should use at least the sequence of objects to construct the sequence of actions.

It should be noted that, the use of the user data according to the embodiments of the present invention is performed on the premise of user authorization, for example, in the case of user authorization, confirmation or active selection by the user, the user data is used.

Step S120, determining, by a processor, behavior characteristics of a target object to be recommended, the behavior characteristics corresponding to the behavior sequence, wherein the behavior characteristics at least represent the matching degree of the target object and the behavior sequence.

The target object can be any object to be recommended by the online internet platform, and when the target object is recommended to the target user, the recommendation ordering of the target object is determined in a personalized way according to the condition of the target user, so that the recommendation ordering of the target object is determined in a personalized way according to the user.

In the embodiment of the invention, aiming at the target object, the embodiment of the invention can extract the corresponding behavior characteristics of the behavior sequences of the target object and the target user so as to embody the matching degree of the behavior sequences of the target object and the target user; in an alternative implementation, the behavior sequence may include an object sequence corresponding to an object of the target user performing the set behavior, and may further include an object type sequence corresponding to an object of the target user performing the set behavior.

In an alternative implementation, the embodiment of the present invention may use a transducer to extract the attention score of the target object corresponding to the behavior sequence, so as to obtain the behavior feature; it should be noted that, the transformer is a feature extractor based on an attention mechanism, and the embodiment of the invention can score a target object to be recommended based on a behavior sequence of a target user to form an attention score, so that the attention score is used as the behavior feature.

Step S130, determining a recommended ranking value of the target object for the target user at least according to the behavior characteristics through a processor; the recommendation ordering value is related to a recommendation ordering of the target object for the target user.

Based on step S120, the embodiment of the invention can determine the behavior characteristics reflecting the matching degree of the target object to be recommended and the behavior sequence on the basis of reflecting the object favored by the target user on the online internet platform through the behavior sequence; further in step S130, according to at least the behavior characteristics, the embodiment of the present invention may determine a recommended ranking value of the target object for the target user; because the recommendation ordering value is related to the recommendation ordering of the target object for the target user, the recommendation ordering value can be used for determining the recommendation ordering of the target object for the target user, so that the determined recommendation ordering of the target object for the target user can be matched with the preference degree of the target user, the recommendation ordering of the target user for the preferential object of the online interconnection platform can be ensured to be in front, and the target recommendation ordering for the user individuation can be determined.

In an alternative implementation, the recommendation-ranking value may be positively correlated with the recommendation-ranking of the target object, e.g., the higher the recommendation-ranking value, the earlier the recommendation-ranking of the target object.

Optionally, further, the embodiment of the present invention may recommend the target object to the target user according to the recommendation rank corresponding to the recommendation rank value.

Compared with a mode of determining object recommendation ordering by simply utilizing the statistical characteristics of the objects, the method and the device can determine the object recommendation ordering based on the objects favored by a single user on an online internet platform, achieve personalized determination of the object recommendation ordering for the user, and provide a basis for improving the accuracy of object recommendation.

In an optional implementation, in order to further improve the accuracy of the recommendation ordering of the object determined for the user, the embodiment of the present invention may further determine the discretization feature at least through the statistical feature of the target object to be recommended, so as to combine the discretization feature with the behavior feature determined in step S120 shown in fig. 1 to determine the recommendation ordering of the target object for the target user; optionally, fig. 2 shows another flowchart of a recommendation ordering determining method provided by an embodiment of the present invention, where the method may be applied to a server of an online internet platform, as shown in fig. 2, and the flowchart may include:

step S200, obtaining a behavior log of the target user.

Step S210, analyzing the behavior log through a processor, and determining a behavior sequence of the target user, wherein the behavior sequence at least comprises an object sequence corresponding to an object of the target user for executing the set behavior and an object type sequence corresponding to the object of the target user for executing the set behavior; wherein the objects in the object sequence are arranged according to the time sequence of the target user executing the set behavior.

In step S210, the behavior sequence includes, in addition to an object sequence of an object reflecting real-time preference of the target user within a certain time, an object type sequence corresponding to an object for the target user to execute a set behavior, so that an object favored by the target user in a long-tail object in an online internet platform can be covered by the object type sequence; in the case that the behavior sequence includes the object sequence and the object type sequence, the behavior sequence can reflect the object favored by the target user on the online internet platform from multiple angles (the angle that the user prefers the object in real time, the angle that the object type corresponds to the favored object, and the like).

Of course, the action sequence is only optional, and the effect that the action sequence reflects the object favored by the target user on the online internet platform can be achieved on the basis that the action sequence at least comprises the object sequence.

Step S220, merging vectors corresponding to the target object to be recommended with vectors corresponding to the behavior sequence through a processor, splitting the merged vectors into a plurality of blocks of vectors, and extracting information of each split block of vectors to obtain attention scores of the target object for the behavior sequence, wherein the attention scores are used as the behavior characteristics.

Alternatively, step S220 may be considered as an alternative implementation of step S120 shown in fig. 1, and step S220 may be implemented by using a transducer; in an alternative implementation, the transducer may combine the vector corresponding to the target object with the vector corresponding to the behavior sequence, so as to split the combined vector into a plurality of blocks of vectors, and because the information emphasis of each split block of vector is different, the embodiment of the invention may extract the information of each split block of vector, so as to form the attention score, and obtain the behavior characteristics of the target object corresponding to the behavior sequence; in an example, assuming that the vector of the target object is 10 dimensions and the vector of the behavior sequence is also 10 dimensions, the vector after merging is 20 dimensions, and the embodiment of the present invention may divide the vector after merging into Z blocks (the value of Z may be set), so that each block is equivalent to a dimension, and the emphasis point of the extracted information of each dimension is different, and further extract the information of each block after dividing, the attention score may be obtained, and the attention score may be used as the behavior feature to represent the matching degree of the target object and the behavior sequence.

And step S230, discretizing at least the statistical characteristics of the target object by a processor to obtain discretized characteristics.

The statistical features may refer to the description of the corresponding parts, and represent the overall user preference of the target object in the overall user plane, and take e-commerce, online take-out and other scenes as examples, where the statistical features of the target object may represent the monthly sales volume, the repurchase rate, the number of user clicks in the last certain day, and the like of the target object.

According to the embodiment of the invention, at least the statistical characteristics of the target object can be discretized through the processor so as to obtain discretized characteristics; in an optional further implementation, the statistical features of the target object may also be discretized with at least one of the following features: attribute characteristics of a target user, attribute characteristics of a target object, interaction class characteristics of the target user and recommendation upper and lower class characteristics;

optionally, the attribute features of the target user may be features constructed by user information such as age, gender, geographic location, etc. of the target user; the attribute features of the object can be the features constructed by the object information such as the name of the object, the type of the object, the online time of the object and the like; the interaction characteristics can be interaction characteristics of the user on the online interconnection platform and the object, for example, information representing the click rate of the user on a certain object in the last 30 days and the like; the recommendation context feature indicates information such as weather and date when the recommendation is made.

In an optional implementation of step S230, the embodiment of the present invention may perform discretization processing on the attribute feature of the target user, the attribute feature of the target object, the statistical feature of the target object, the interaction class feature of the target user, and the recommended context class feature, so as to obtain a discretized feature.

The more comprehensive the information contained by the discretization features, the more favorable the accuracy of the recommendation ordering determined later is improved.

And step S240, determining a recommended ranking value of the target object for the target user by the processor according to the behavior characteristics and the discretization characteristics.

Based on the behavior characteristics, the embodiment of the invention can further utilize the discretization characteristics to determine the recommended sorting value of the target object for the target user, thereby improving the accuracy of the finally determined recommended sorting.

Optionally, to further refine the useful information in the behavior feature and the discretized feature to improve the accuracy of the determined recommendation-ranking result, an alternative implementation of determining the recommendation-ranking value in step S240 may be as shown in fig. 3, including:

step S300, extracting first-order and second-order characteristic information of the discretization characteristic.

Optionally, for the discretized feature, the embodiment of the invention can extract the first-order and second-order feature information for the discretized feature by using an FM (Factorization Machine, factorizer), and the FM is mainly used for predicting the association between different features in the recommendation system, and is particularly suitable for recommendation of sparse data.

And step S310, combining the discretization features with the behavior features, and extracting hidden abstract features of the combined features.

The embodiment of the invention can combine the discretized features with the behavior features, and in an alternative implementation, hidden abstract features of the combined features can be further extracted through DNN (Deep Neural Networks, deep neural network).

And step 320, connecting the hidden layer abstract feature and the first-order and second-order feature information to obtain a connected feature.

The embodiment of the invention can connect the hidden abstract feature with the first-order and second-order feature information to obtain the connected feature.

Step S330, determining a recommended sorting value corresponding to the coupled features.

In an alternative implementation, the embodiment of the present invention may output, using the activation layer, a recommended ranking value corresponding to the feature after the connection, where the recommended ranking value may be considered as the recommended ranking value of the target object determined in step S130. Based on the flow shown in fig. 3, the embodiment of the invention can extract the useful information of the discretization feature and the useful information after the combination of the discretization feature and the behavior feature, so as to determine the recommended ranking value of the target object, thereby improving the accuracy of the determined recommended ranking value.

In an alternative implementation, the embodiment of the present invention may determine the recommended ranking value of the target object for the target user by using an algorithm model, as shown in fig. 4, where the structure of the algorithm model may include: feature extraction layers (e.g., transfomers), discretization layers, factorizers, binding layers, DNN layers, tie layers, and activation layers; in an alternative implementation, an activation layer, such as a softmax layer, is widely used in multi-class scenarios by mapping the inputs to real numbers between 0-1, and normalizing the guaranteed sum to 1.

In combination with the algorithm model shown in fig. 4, the implementation process of the recommendation ordering determining method provided by the embodiment of the present invention may be shown in fig. 5, and includes the following processes:

extracting attention scores corresponding to the behavior sequences of the target object to be recommended and the target user by utilizing a feature extraction layer so as to obtain behavior features;

performing discretization processing on the attribute characteristics of the target user, the attribute characteristics of the target object, the statistical characteristics of the target object, the interactive class characteristics of the target user and the recommended context class characteristics by utilizing the discretization layer to obtain discretization characteristics;

acquiring first-order and second-order characteristic information of the discretization characteristic by using a factor decomposition machine;

Combining the discretized features with the behavioral features using a combining layer to obtain combined features;

extracting hidden layer abstract features of the combined features by using a DNN layer;

connecting the hidden layer abstract feature with the first-order and second-order feature information by using a connecting layer to obtain a connected feature;

and inputting the connected features into an activation layer, and outputting a recommended sorting value of the target object for the target user by the activation layer.

The embodiment of the invention utilizes the algorithm model to realize the processing of the data related to the embodiment of the invention, can change the process of determining the recommendation ordering into the execution process of the algorithm model, can conveniently, efficiently and accurately realize the determination of the object recommendation ordering aiming at the individuation of the user, and improves the accuracy of object recommendation.

Optionally, as described in the foregoing embodiments, it may be seen that in the embodiments of the present invention, the recommended ranking value of the target object for the target user may be expressed as:

p＝Softmax(fm(x1)，dnn(x1，x2))

in the above formula, x1 represents first-order and second-order feature information of the discretization feature acquired by the factoring machine, it should be noted that x1 covers attribute features of the target user, attribute features of the target object, statistical features of the target object, contextual features and the like, and because FM and DNN perform better on the discretization feature, the statistical features in x1 are discrete features after equal-frequency discretization; x2 represents a behavior feature, and the embodiment of the invention obtains the behavior feature by scoring attention scores of target objects to be recommended through a transducer based on a behavior sequence; x2 and x1 are combined with first-order and second-order features obtained by a factor decomposition machine after hidden layer abstract features are extracted through a DNN layer, and then a recommended sorting value of a target object for a target user is output through a softmax layer, wherein the recommended sorting value can be a numerical value between 0 and 1; in an alternative implementation, the higher the recommendation ranking value, the earlier the recommendation ranking.

In an alternative implementation, the sequence of objects in the behavior sequence of the target user may be extracted by an online dotting log of the target user, while the sequence of object types in the behavior sequence may be extracted by an offline log of the target user; as shown in fig. 6, the behavior log of the target user may include an online dotting log of the target user and an offline log of the target user;

according to the embodiment of the invention, the objects of the target user, which are ordered according to the time sequence and execute the set behaviors, can be determined from the online dotting log of the target user, so that an object sequence is formed; optionally, the object may be represented by an object identifier, and different objects have different object identifiers, so that the object sequence may be a sequence of object identifiers, that is, a sequence formed by object identifiers corresponding to objects arranged in time sequence, in which the target user performs the set behavior, as the object sequence; by way of example, the object identification may be an object ID or the like;

in an alternative example, the embodiment of the present invention may determine, from an online dotting log of a target user, a sequence of objects that the target user has clicked last a first time, a sequence of objects that have purchased last a second time, and a sequence of objects that have purchased last a third time, thereby forming the object sequence, that is, the object sequence of the target user may include a sequence of objects that have clicked last a first time, a sequence of objects that have purchased last a second time, and a sequence of objects that have purchased last a third time, which are arranged in chronological order; optionally, specific values of the first times, the second times and the third times may be set according to actual situations;

The object sequence is more concerned with the object interacted with the target user, and the occurrence probability of some popular objects is much larger than that of some cold objects and newly online objects (namely long-tail objects), so that the object sequence is very likely to be incapable of recording the object favored by the target user in the long-tail objects, and therefore, the embodiment of the invention introduces the object type sequence; the embodiment of the invention can determine the sequence of the object types corresponding to the object of the user executing the set behavior from the offline log of the target user to form an object type sequence; for example, in an online take-away scene, the embodiment of the invention can introduce a vegetable class sequence corresponding to a vegetable purchased by a user, so that a plurality of objects which are cold or newly put on shelf can be covered, related information of user behaviors and long-tail objects can be mined, and the recommendation accuracy is improved;

in one example, the embodiment of the invention can select an offline ordering log of a target user with t+1 (the t value can be set), determine the object type corresponding to the user ordering object, and form an object type sequence.

Therefore, the embodiment of the invention can mine the real-time favorite objects of the user through the object sequence and mine the object types corresponding to the favorite objects of the user through the object type sequence, so as to realize covering the favorite objects of the user in the long-tail object, and further improve the recommendation accuracy on the basis of realizing personalized determination of recommendation ordering for the user.

In an optional application example, the recommendation ordering determining method provided by the embodiment of the invention can be used for realizing the recommendation of the commercial tenant or commodity of the online take-out platform personalized by the user, and taking the personalized recommendation of the commodity for the user as an example, the application process provided by the embodiment of the invention can be realized as follows, and the recommendation process of the commercial tenant is realized in the same way.

Extracting commodity sequences clicked by a user for N times recently, commodity sequences ordered for M times recently and commodity sequences purchased again for P times recently from an online dotting log of the user on an online take-out platform according to a time sequence to form commodity sequences of target users;

because the online dotting daily records that the commodity sequence clicked/purchased again/ordered by the user occupies a large memory and cannot be added with the sequence information of commodity types, the embodiment of the invention can determine the commodity type sequence corresponding to the commodity ordered offline in the last t+1 day of the user from the data of the offline ordered in the last t+1 day of the user; alternatively, the values of N, M, P and t may be set;

thus, the sequence of goods and the sequence of goods types may form a sequence of actions for the user;

further, the embodiment of the invention can acquire the attribute characteristics of the user, the attribute characteristics of the commodity to be recommended, the statistical characteristics of the commodity to be recommended, the interactive characteristics of the user and the recommended context characteristics as basic data for forming the discretization characteristics;

Determining a recommendation ordering value of the commodity to be recommended for the user based on the behavior sequence and basic data forming discretized features by using the algorithm model shown in fig. 4 and the process shown in fig. 5;

furthermore, based on the recommendation ordering value of the commodity, the embodiment of the invention can determine the recommendation ordering of the commodity for the user, and optionally, the higher the recommendation ordering value of the commodity is, the higher the recommendation ordering of the commodity for the user is.

The recommendation ordering determining method provided by the embodiment of the invention can determine the object recommendation ordering aiming at the individuation of the user, and provides a basis for improving the accuracy of object recommendation.

The foregoing describes several embodiments of the present invention, and the various alternatives presented by the various embodiments may be combined, cross-referenced, with each other without conflict, extending beyond what is possible embodiments, all of which are considered to be embodiments of the present invention disclosed and disclosed.

The following describes a recommendation order determining device provided in the embodiment of the present invention, where the device described below may be considered as a functional module required to be set by a server to implement the recommendation order determining manner provided in the embodiment of the present invention. The apparatus contents described below may be referred to in correspondence with the method contents described above.

In an alternative implementation, fig. 7 shows a block diagram of a recommendation ordering determining apparatus provided in an embodiment of the present invention, and as shown in fig. 7, the apparatus may include:

the log obtaining module 100 is configured to obtain a behavior log of a target user;

a sequence determining module 110, configured to parse the behavior log through a processor, determine a behavior sequence of the target user, where the behavior sequence at least includes an object sequence corresponding to an object of the target user performing the set behavior, and objects in the object sequence are arranged according to a time sequence of the target user performing the set behavior;

a behavior feature determining module 120, configured to determine, by using a processor, a behavior feature corresponding to the behavior sequence of the target object to be recommended, where the behavior feature at least represents a matching degree between the target object and the behavior sequence;

a ranking determining module 130, configured to determine, by a processor, a recommended ranking value of the target object for the target user based at least on the behavior feature; the recommendation ordering value is related to a recommendation ordering of the target object for the target user.

Optionally, the behavior sequence further includes an object type sequence; the object type sequence comprises an object type corresponding to an object of which the target user executes the set behavior.

Optionally, the behavior feature determining module 120 is configured to determine, by the processor, a behavior feature of the target object to be recommended corresponding to the behavior sequence, where the determining module includes:

Optionally, the behavior feature determining module 120 is configured to determine, by the processor, an attention score of the target object for the behavior sequence, including:

combining vectors corresponding to the target object and vectors corresponding to the behavior sequence through a processor;

dividing the combined vector into a plurality of blocks of vectors, and extracting the information of each divided block of vectors to obtain the attention score of the target object aiming at the behavior sequence.

Optionally, the attention score in the embodiment of the present invention is determined by using a feature extraction layer.

Optionally, the ranking determining module 130 is configured to determine, by a processor, a recommended ranking value of the target object for the target user according to at least the behavior feature, including:

Optionally, the ranking determining module 130 is configured to perform, by the processor, discretizing at least the statistical feature of the target object to obtain a discretized feature, where the discretized feature includes:

Alternatively, the discretizing process described in the embodiments of the present invention may be performed using a discretizing layer.

Optionally, the ranking determining module 130 is configured to determine, by the processor, a recommended ranking value of the target object for the target user according to the behavior feature and the discretized feature, where the determining module includes:

Optionally, in the embodiment of the present invention, the first-order and second-order feature information is extracted by using a factorizer; extracting the hidden layer abstract features by using a deep neural network layer; the hidden layer abstract feature and the first-order and second-order feature information are connected by using a connecting layer; and determining the recommended sorting value corresponding to the connected features by using an activation layer.

Optionally, the sequence determining module 110 is configured to determine, by analyzing, by the processor, the behavior sequence of the target user, and includes:

Optionally, fig. 8 shows another block diagram of a recommendation ordering determining apparatus provided in an embodiment of the present invention, and in combination with fig. 7 and fig. 8, the apparatus may further include:

and the recommending module 140 is configured to recommend the target object to the target user according to the recommendation sequence corresponding to the recommendation sequence value.

The server provided by the embodiment of the invention can realize the recommendation ordering determining method provided by the embodiment of the invention through the recommendation ordering determining device in the form of a loading program; in an alternative implementation, fig. 9 shows an alternative hardware structure of a server provided by an embodiment of the present invention, and referring to fig. 9, the server may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

in the embodiment of the invention, the number of the processor 1, the communication interface 2, the memory 3 and the communication bus 4 is at least one, and the processor 1, the communication interface 2 and the memory 3 complete the communication with each other through the communication bus 4;

alternatively, the communication interface 2 may be an interface of a communication module for performing network communication;

the processor 1 may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

The memory 3 may comprise a high-speed RAM memory or may further comprise a non-volatile memory, such as at least one disk memory.

Wherein the memory 3 stores one or more computer-executable instructions that the processor 1 invokes to perform the recommendation ordering determining method provided by the embodiments of the present invention.

The embodiment of the invention also provides a storage medium which can store one or more computer executable instructions for executing the recommendation ordering determining method provided by the embodiment of the invention.

Although the embodiments of the present invention are disclosed above, the present invention is not limited thereto. Various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the invention, and the scope of the invention should be assessed accordingly to that of the appended claims.

In summary, it can be seen that in the 1 st aspect, the embodiment of the present invention provides a recommendation ordering determining method, which includes:

acquiring a behavior log of a target user;

With reference to aspect 1, in a 1 st implementation manner of aspect 1, the behavior sequence further includes an object type sequence; the object type sequence comprises an object type corresponding to an object of which the target user executes the set behavior.

With reference to the 1 st aspect or the 1 st implementation manner of the 1 st aspect, in a 2 nd implementation manner of the 1 st aspect, the determining, by the processor, a behavior feature of the target object to be recommended corresponding to the behavior sequence includes:

With reference to implementation manner of the 2 nd aspect of the 1 st aspect, in implementation manner of the 3 rd aspect of the embodiment of the present invention, the determining, by the processor, an attention score of the target object for the behavior sequence includes:

With reference to implementation manner of 3 rd aspect of the present invention, in implementation manner of 4 th aspect of 1, the attention score is determined by using a feature extraction layer.

With reference to the 1 st aspect or the 1 st implementation manner of the 1 st aspect, in a 5 th implementation manner of the 1 st aspect, the determining, by the processor, a recommended ranking value of the target object for the target user according to at least the behavior feature includes:

With reference to the 5 th implementation manner of the 1 st aspect, in a 6 th implementation manner of the 1 st aspect, the performing, by the processor, discretizing at least the statistical feature of the target object to obtain a discretized feature includes:

With reference to the 5 th or 6 th implementation manner of the 1 st aspect, in a 7 th implementation manner of the 1 st aspect, the discretization processing is performed by using a discretization layer.

With reference to the 5 th implementation manner of the 1 st aspect, in an 8 th implementation manner of the 1 st aspect, the determining, by the processor, a recommended ranking value of the target object for the target user according to the behavior feature and the discretization feature includes:

With reference to the 8 th implementation manner of the 1 st aspect, in a 9 th implementation manner of the 1 st aspect, the first-order and second-order feature information is extracted by using a factorizer; extracting the hidden layer abstract features by using a deep neural network layer; the hidden layer abstract feature and the first-order and second-order feature information are connected by using a connecting layer; and determining the recommended sorting value corresponding to the connected features by using an activation layer.

With reference to implementation manner of the 1 st aspect, in a 10 th implementation manner of the 1 st aspect, the analyzing, by the processor, the behavior log, and determining a behavior sequence of the target user include:

With reference to aspect 1, in an 11 th implementation manner of aspect 1, the method further includes:

and recommending the target object to the target user according to the recommendation sequence corresponding to the recommendation sequence value.

In the 2 nd aspect, an embodiment of the present invention provides a recommendation ordering determining device, including:

With reference to aspect 2, in a 1 st implementation manner of aspect 2, the behavior sequence further includes an object type sequence; the object type sequence comprises an object type corresponding to an object of which the target user executes the set behavior.

With reference to the 2 nd aspect or the 1 st implementation manner of the 2 nd aspect, in a 2 nd implementation manner of the embodiment of the present invention, the behavior feature determining module, configured to determine, by using a processor, a behavior feature of a target object to be recommended corresponding to the behavior sequence includes:

With reference to implementation manner of the 2 nd aspect, in a 3 rd implementation manner of the 2 nd aspect, the determining, by the processor, the attention score of the target object for the behavior sequence includes:

With reference to implementation manner of 3 rd aspect of 2, in implementation manner of 4 th aspect of 2, the attention score is determined by using a feature extraction layer.

With reference to

aspect

2 or 1 st implementation manner of aspect 2, in a 5 th implementation manner of aspect 2, the determining, by a processor, a recommended ranking value of the target object for the target user at least according to the behavior feature includes:

With reference to the 5 th implementation manner of the 2 th aspect, in a 6 th implementation manner of the 2 th aspect, the ranking determining module is configured to perform, by using the processor, discretizing at least the statistical feature of the target object to obtain a discretized feature, where the discretizing feature includes:

With reference to the 5 th or 6 th implementation manner of the 2 nd aspect, in a 7 th implementation manner of the 2 nd aspect, the discretizing is performed by using a discretization layer.

With reference to the 5 th implementation manner of the 2 th aspect, in an 8 th implementation manner of the 2 nd aspect, the determining, by the processor, a recommended ranking value of the target object for the target user according to the behavior feature and the discretization feature includes:

With reference to the 8 th implementation manner of the 2 nd aspect, in a 9 th implementation manner of the 2 nd aspect, the first-order and second-order feature information is extracted by using a factorizer; extracting the hidden layer abstract features by using a deep neural network layer; the hidden layer abstract feature and the first-order and second-order feature information are connected by using a connecting layer; and determining the recommended sorting value corresponding to the connected features by using an activation layer.

With reference to the 1 st implementation manner of the 2 st aspect, in a 10 th implementation manner of the 2 nd aspect, the sequence determining module is configured to parse the behavior log by using a processor, and determining a behavior sequence of the target user includes:

With reference to aspect 2, in an 11 th implementation manner of aspect 2, the apparatus further includes:

and the recommending module is used for recommending the target object to the target user according to the recommending sequence corresponding to the recommending sequence value.

In aspect 3, an embodiment of the present invention provides a server, including at least one memory storing one or more computer-executable instructions and at least one processor invoking the one or more computer-executable instructions to perform the recommendation ordering determining method according to any of the above.

In a 4 th aspect, embodiments of the present invention provide a storage medium storing one or more computer-executable instructions for performing the recommendation ordering determining method of any of the above.

Claims

1. A recommendation ordering determining method, comprising:

acquiring a behavior log of a target user;

determining, by a processor, a recommended ranking value of the target object for the target user based at least on the behavioral characteristics; the recommendation ordering value is related to the recommendation ordering of the target object for the target user;

the determining, by the processor, a recommended ranking value of the target object for the target user based at least on the behavioral characteristics includes:

determining, by the processor, a recommended ranking value of a target object for the target user according to the behavioral characteristics and the discretization characteristics;

the determining, by the processor, a recommended ranking value of a target object for the target user according to the behavioral characteristics and the discretization characteristics includes:

2. The recommendation-ranking-determining method according to claim 1, wherein the behavior sequence further comprises a sequence of object types; the object type sequence comprises an object type corresponding to an object of which the target user executes the set behavior.

3. The recommendation ordering determining method according to claim 1 or 2, wherein the determining, by a processor, a behavior feature of a target object to be recommended corresponding to the behavior sequence comprises:

4. The recommendation ordering determining method according to claim 3, wherein the determining, by the processor, an attention score of the target object for the sequence of actions comprises:

5. The recommendation ordering determining method according to claim 4, wherein the attention score is determined using a feature extraction layer.

6. The recommendation ordering determining method according to claim 1, wherein discretizing, by the processor, at least the statistical features of the target object to obtain discretized features comprises:

7. The recommendation ordering determining method according to claim 1 or 6, wherein the discretization processing is performed with a discretization layer.

8. The recommendation ordering determining method according to claim 1, wherein the first and second order feature information is extracted by a factorizer; extracting the hidden layer abstract features by using a deep neural network layer; the hidden layer abstract feature and the first-order and second-order feature information are connected by using a connecting layer; and determining the recommended sorting value corresponding to the connected features by using an activation layer.

9. The recommendation ordering determining method according to claim 2, wherein the determining the behavior sequence of the target user by the processor parsing the behavior log comprises:

10. The recommendation ordering determining method according to claim 1, further comprising:

11. A recommendation order determining device, comprising:

the ranking determining module is used for determining a recommended ranking value of the target object for the target user at least according to the behavior characteristics through the processor; the recommendation ordering value is related to the recommendation ordering of the target object for the target user;

the ranking determining module, configured to determine, by a processor, a recommended ranking value of the target object for the target user based at least on the behavioral characteristics, includes:

the ranking determining module, configured to determine, by the processor, a recommended ranking value of a target object for the target user according to the behavior feature and the discretized feature, includes:

12. The recommendation ordering determining device according to claim 11, wherein the sequence of actions further comprises a sequence of object types; the object type sequence comprises an object type corresponding to an object of which the target user executes the set behavior.

13. The recommendation ordering determining device according to claim 11 or 12, wherein the behavioral characteristics determining module for determining, by a processor, behavioral characteristics of a target object to be recommended corresponding to the behavioral sequence comprises:

14. The recommendation ordering determining device according to claim 13, wherein the behavioral characteristics determining module for determining, by the processor, an attention score of the target object for the behavioral sequence comprises:

15. The recommendation ordering determining device according to claim 14, wherein the attention score is determined using a feature extraction layer.

16. The recommendation ranking determining apparatus according to claim 11 wherein the ranking determining module for discretizing at least the statistical features of the target object by the processor to obtain discretized features comprises:

17. The recommendation ordering determining device according to claim 11 or 16, wherein the discretization process is performed with a discretization layer.

18. The recommendation rank determining device according to claim 11, wherein the first and second order feature information is extracted using a factorizer; extracting the hidden layer abstract features by using a deep neural network layer; the hidden layer abstract feature and the first-order and second-order feature information are connected by using a connecting layer; and determining the recommended sorting value corresponding to the connected features by using an activation layer.

19. The recommendation ordering determining device according to claim 12, wherein the sequence determining module for parsing the behavior log by a processor, determining a sequence of behaviors of the target user comprises:

20. The recommendation ordering determining device according to claim 11, further comprising:

21. A server comprising at least one memory storing one or more computer-executable instructions and at least one processor invoking the one or more computer-executable instructions to perform the recommendation ordering determining method of any of claims 1-10.

22. A storage medium storing one or more computer-executable instructions for performing the recommendation ordering determination method of any of claims 1-10.