CN110569428B

CN110569428B - Recommendation model construction method, device and equipment

Info

Publication number: CN110569428B
Application number: CN201910731153.6A
Authority: CN
Inventors: 胡丁相; 操颖平
Original assignee: Advanced New Technologies Co Ltd
Current assignee: Advanced New Technologies Co Ltd; Advantageous New Technologies Co Ltd
Priority date: 2019-08-08
Filing date: 2019-08-08
Publication date: 2023-10-20
Anticipated expiration: 2039-08-08
Also published as: CN110569428A

Abstract

The embodiment of the specification discloses a recommendation model construction method, device and equipment, wherein the method comprises the following steps: acquiring target scene information for information recommendation; if the scene corresponding to the target scene information is contained in a preset scene set, acquiring a feature class corresponding to the scene set, and acquiring sample data corresponding to the target scene information according to the feature class; based on the feature category, extracting the features of the sample data to obtain target features corresponding to the sample data; training the recommendation model corresponding to the scene set by using the target features to obtain the information recommendation model corresponding to the target scene information.

Description

Recommendation model construction method, device and equipment

Technical Field

The present document relates to the field of computer technologies, and in particular, to a method, an apparatus, and a device for constructing a recommendation model.

Background

With the continuous progress of networking and the increasing abundance of various products and information contents, how to push information such as appropriate products or information contents to users in appropriate scenes becomes the difficulty and pain of the operation of the products or information contents.

In order to push information such as proper products or information content and the like to a user under a proper scene, the information can be realized through a personalized recommendation scheme of big data and related algorithms, specifically, when a requirement of information recommendation is acquired, a corresponding model is required to be built according to the requirement or the scene of information recommendation, and a corresponding link is built for the built model and a corresponding external system, so that a result output through the model can be provided for the corresponding external system for subsequent processing. However, in the personalized recommendation scheme based on big data and related algorithms, the corresponding model and the corresponding link are required to be reconstructed each time the information recommendation requirement is acquired, so that a long time is required for online of the information recommendation scheme, the constructed model is difficult to mass produce, in addition, in consideration of the fact that the information recommendation scene is often more, the corresponding model and the link are required to be reconstructed each time the new information recommendation requirement is acquired, fragmentation is serious, and therefore, a general marketing recommendation model framework is needed to be provided, and the online efficiency of the model is improved.

Disclosure of Invention

The embodiment of the specification aims to provide a recommendation model construction method, device and equipment, so as to provide a general marketing recommendation model framework, and further improve the online efficiency of the model.

In order to achieve the above technical solution, the embodiments of the present specification are implemented as follows:

the embodiment of the specification provides a method for constructing a recommendation model, which comprises the following steps:

acquiring target scene information for information recommendation;

if the scene corresponding to the target scene information is contained in a preset scene set, acquiring a feature class corresponding to the scene set, and acquiring sample data corresponding to the target scene information according to the feature class;

based on the feature category, extracting the features of the sample data to obtain target features corresponding to the sample data;

training the recommendation model corresponding to the scene set by using the target features to obtain the information recommendation model corresponding to the target scene information.

Optionally, the feature class includes one or more of a user feature class, a feature class corresponding to the recommended information, and a behavior feature class of the user on the recommended information.

Optionally, the obtaining, according to the feature class, sample data corresponding to the target scene information includes:

determining a data embedding point in recommended information corresponding to the target scene information according to the characteristic category;

And taking the data obtained based on the data embedded points as sample data corresponding to the target scene information.

Optionally, the method further comprises:

according to the feature types contained in the target features, carrying out corresponding feature extraction processing on the target features to obtain extracted features;

training the recommendation model corresponding to the scene set by using the target features to obtain an information recommendation model corresponding to the target scene information, wherein the training comprises the following steps:

training the recommendation model corresponding to the scene set by using the extracted features to obtain the information recommendation model corresponding to the target scene information.

Optionally, the performing, according to the feature type included in the target feature, a corresponding feature extraction process on the target feature to obtain an extracted feature includes:

if the target feature comprises character type features, discretizing the character type features in the target feature to obtain first features, and taking the first features as the extracted features;

if the target feature comprises a numerical feature, carrying out box division processing and normalization processing on the numerical feature in the target feature to obtain a second feature, and taking the second feature as the extracted feature;

And if the target feature comprises the predefined feature, carrying out Cartesian product cross extraction processing on the predefined feature in the target feature to obtain a third feature, and taking the third feature as the extracted feature.

Optionally, the recommendation model is a deep learning model.

Optionally, the recommended model is a Wide & Deep model, a depth factorizer Deep fm model, an attention factorizer AFM model, or a depth interest network DIN model.

Optionally, in the training the recommendation model corresponding to the target scene information using the target feature, the method further includes:

debugging the recommended model based on a preset single NoteBook debugging mechanism; and/or the number of the groups of groups,

the recommendation model is trained based on a predetermined Yarn distributed training mechanism.

The embodiment of the specification provides a device for constructing a recommendation model, which comprises:

the scene information acquisition module is used for acquiring target scene information for information recommendation;

the sample acquisition module is used for acquiring a feature class corresponding to a scene set if the scene corresponding to the target scene information is contained in a preset scene set, and acquiring sample data corresponding to the target scene information according to the feature class;

The feature extraction module is used for carrying out feature extraction on the sample data based on the feature category to obtain target features corresponding to the sample data;

and the training module is used for training the recommendation model corresponding to the scene set by using the target features to obtain the information recommendation model corresponding to the target scene information.

Optionally, the sample acquisition module includes:

the data embedding point determining unit is used for determining the data embedding point in the recommended information corresponding to the target scene information according to the characteristic category;

and the sample acquisition unit is used for taking the data obtained based on the data embedded point as sample data corresponding to the target scene information.

Optionally, the apparatus further comprises:

the feature extraction module is used for carrying out corresponding feature extraction processing on the target features according to the feature types contained in the target features to obtain extracted features;

and the training module is used for training the recommendation model corresponding to the scene set by using the extracted features to obtain the information recommendation model corresponding to the target scene information.

Optionally, the feature extraction module includes:

The first feature extraction unit is used for carrying out discretization processing on the character type features in the target features if the character type features are included in the target features to obtain first features, and taking the first features as the extracted features;

the second feature extraction unit is used for carrying out box division processing and normalization processing on the numerical type features in the target features if the numerical type features are included in the target features to obtain second features, and taking the second features as the extracted features;

and the third feature extraction unit is used for carrying out Cartesian product cross extraction processing on the predefined features in the target features if the predefined features are included in the target features to obtain third features, and taking the third features as the extracted features.

Optionally, the recommendation model is a deep learning model.

The embodiment of the specification provides a recommendation model construction device, the recommendation model construction device includes:

a processor; and

a memory arranged to store computer executable instructions that, when executed, cause the processor to:

acquiring target scene information for information recommendation;

As can be seen from the technical solutions provided in the embodiments of the present disclosure, by obtaining target scene information for information recommendation, if a scene corresponding to the target scene information is included in a predetermined scene set, sample data corresponding to the target scene information is obtained according to a feature class corresponding to the predetermined scene set, and further, target features corresponding to the sample data are obtained, and a recommendation model corresponding to the predetermined scene set is trained by using the target features, so that an information recommendation model corresponding to the target scene information is obtained, and thus, by performing standardized processing on features, sample data and the recommendation model corresponding to the predetermined scene set in advance, a general marketing recommendation model frame is provided for each scene (a scene corresponding to the target scene information) in the predetermined scene set, so that each scene in the predetermined scene set can multiplex the construction mode and flow of the same set of recommendation model, thereby simplifying the construction flow of the recommendation model and greatly saving the online efficiency of the model.

Drawings

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

FIG. 1 is a schematic diagram of an embodiment of a method for constructing a recommendation model;

FIG. 2 is a schematic diagram of a page of an information recommendation system according to the present disclosure;

FIG. 3 is a schematic diagram illustrating another embodiment of a method for constructing a recommendation model according to the present disclosure;

FIG. 4 is a schematic diagram of an embodiment of a recommendation model construction apparatus according to the present disclosure;

FIG. 5 is an embodiment of a recommendation model building apparatus according to the present disclosure.

Detailed Description

The embodiment of the specification provides a recommendation model construction method, device and equipment.

In order to make the technical solutions in the present specification better understood by those skilled in the art, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only some embodiments of the present specification, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.

Example 1

As shown in fig. 1, the embodiment of the present disclosure provides a method for constructing a recommendation model, where an execution body of the method may be a server, where the server may be an independent server, or may be a server cluster formed by a plurality of servers, and the server may be a background server of a website (such as an online shopping website or a shopping application), or may be a server of a service (such as an information recommendation service), or the like. The method can be used for providing a general marketing recommendation model framework so as to improve the online efficiency of the model and other processes. The method specifically comprises the following steps:

in step S102, target scene information for information recommendation is acquired.

The target scene information may be any scene information that needs to be recommended by information, such as information of a scene that is recommended by a product (e.g., an insurance product or a commodity), scene information that is recommended by rights information (e.g., a coupon, etc.), scene information that is recommended by a content (e.g., a document, etc.), or scene information that is recommended by a marketing activity (e.g., a annual celebration activity of a store, etc.), etc.

In implementation, with the continuous progress of networking and the increasing abundance of various products and information contents, how to push information such as appropriate products or information contents to users in appropriate scenes becomes the difficulty and pain of the operation of the products or information contents. In order to push information such as proper products or information contents for users in proper scenes, corresponding users can be selected for the information to be recommended in a manual rule mode, however, in the mode, how to achieve accurate matching of the information to be recommended and the users in massive information recommendation scenes becomes an important problem to be solved urgently, and in practical application, in massive information recommendation scenes, accurate matching of the information to be recommended and the users is generally difficult to achieve. Because the above-mentioned problems generally exist in the above-mentioned manner, personalized recommendation schemes based on big data and related algorithms are generated, and compared with the information recommendation schemes based on the above-mentioned manual rules, the information recommendation schemes based on big data and related algorithms can bring better user experience, and in addition, the business effects (such as click rate and conversion rate) are obviously improved, and meanwhile, the operation efficiency is also greatly improved. The personalized recommendation scheme based on big data and related algorithm specifically comprises the following steps: when the requirement of information recommendation is acquired, a corresponding model is required to be built according to the requirement or the scene of the information recommendation, and a corresponding link is built for the built model and a corresponding external system, so that the result output through the model can be provided for the corresponding external system for subsequent processing. However, in the personalized recommendation scheme based on big data and related algorithms, a corresponding model needs to be reconstructed each time the requirement of information recommendation is acquired, and a link for connecting the model with an external system is needed, and in the process of constructing the corresponding model, the period of a model iteration process is long, and the corresponding model needs to be redeveloped each time, so that the online of the information recommendation scheme needs a long time, the constructed model is difficult to produce in a mass mode, and in the process of constructing the model, experience of technicians is often relied on, and manual rules are often used as the main rule. Based on the above, considering that the information recommendation is often more in scenes and further causes serious fragmentation, and when new information recommendation needs are obtained each time, a corresponding model and a link need to be reconstructed, so that the cost of information recommendation is higher, therefore, a general marketing recommendation model framework needs to be provided, so that the online efficiency of the model is improved. The embodiment of the specification provides a technical scheme capable of improving the online efficiency of a model, which specifically includes the following contents:

When a certain merchant or a provider of a certain product, etc. needs to recommend certain commodity or information to a user, an information recommendation system may be preset, the information recommendation system may provide an information recommendation scheme for the merchant or the provider, etc. and may recommend set information to a corresponding user based on the determined information recommendation scheme, as shown in fig. 2, the information recommendation system may include an input box of an information recommendation requirement, a determination key, a cancel key, etc., the merchant or the provider may send the information recommendation requirement of the merchant or the provider, etc. through the information recommendation system, and the information recommendation requirement may include information required to be recommended to the user, attribute information of the recommended user (such as age, occupation, sex, etc. of the recommended user), information of a scene in which information recommendation is performed (i.e., target scene information), etc. After the server receives the information recommendation requirement, target scene information for information recommendation can be obtained from the information recommendation requirement.

In step S104, if the scene corresponding to the target scene information is included in the predetermined scene set, a feature class corresponding to the predetermined scene set is obtained, and sample data corresponding to the target scene information is obtained according to the feature class.

The predetermined scene set may be a set of a plurality of preset scenes, one or more scenes may be included in the predetermined scene set, and the predetermined scene set may include one or more scenes of a product recommended scene, an information class information recommended scene, a rights information recommended scene, and the like. The feature class may be a class corresponding to one or more features, for example, a user feature class, an information feature class, etc., and may be specifically set according to actual situations, which is not limited in the embodiment of the present disclosure. The sample data may be sample data for constructing a model for information recommendation.

In an implementation, in order to improve the online efficiency of the model, a set of general marketing recommendation model frame may be provided, where the general marketing recommendation model frame may include a set of sample data, a set of features, a set of models, a set of links connected with an external system, and the like, and the general marketing recommendation model frame may be applied to a plurality of different scenes, that is, for a specific scene in a predetermined scene set, sample data, features, models, links, and the like may be respectively constructed for the specific scene based on the set of sample data, the set of features, the set of models, the set of links connected with the external system, and the like. For a set of features, a general feature system based on various different scenes can be constructed, and the general feature system can comprise user features, recommended information features and user behavior features on the recommended information, wherein the user features mainly comprise user portrait features which tend to be static, the user portrait features can be obtained from personal information of a user, for example, corresponding features and the like can be extracted from information such as gender, age, hobbies, occupation and the like in the personal information of the user, and in addition, the user portrait features are features which tend to be static, so that the user portrait features can be multiplexed by various different types of scenes. The features of the recommended information mainly may include basic information class features of the recommended information and flow class features corresponding to the recommended information, where the basic information class features of the recommended information may include features corresponding to information such as a name, an information category, a content, and the like of the recommended information, and the flow class features corresponding to the recommended information may be features for characterizing a degree of interest of a user in the recommended information, for example, features corresponding to the information that the user clicks the recommended information, features corresponding to the information that the user does not click the recommended information, and the like. The behavior characteristics of the user on the recommended information may be mainly related characteristics of the user on the history behavior of the recommended information, for example, characteristics corresponding to the information that the user clicks the recommendation, characteristics corresponding to the information that the user browses the recommendation, and the like. The core of the construction of the universal characteristic system is that the portrait characteristic of the user is more accurate, and simultaneously, the characteristics of the recommended information and the behavior characteristics of the user on the recommended information can be standardized, so that the development of the new recommended information is easy.

The user features may include, in addition to the above user portrait features that tend to be static, features corresponding to the location of the user, the environment in which the user is located, and the like, and may be specifically set according to actual situations, which is not limited in the embodiments of the present disclosure.

For a set of samples, a training sample set formed by splicing a conversion log (which may be a log used for recording recommended information to be clicked or browsed, etc.) and a feature log (which may be a log used for recording other relevant features and the like besides the portrait features of a user) may be constructed, and the core of the samples is standardization of buried points corresponding to the logs, so that a user terminal device is required to record timely response log content at the buried points, specifically, for example, record standardized identification of recommended information, recommended style (or type, etc.) and session identification of recommended information, and meanwhile, record corresponding session identification in the feature log, and correlate the feature log and the conversion log through the session identification to form the training sample set, so that consistency of on-line features and features corresponding to off-line training samples may be ensured.

Based on the above, a plurality of scenes may have a set of sample data, characteristics, a model, and a link connected to an external system in common, and a set of sample data, characteristics, a model, a link connected to an external system, and the like corresponding to different types of scenes may be different, so that when target scene information for information recommendation is acquired, it may be determined to which type of scene the scene corresponding to the target scene information belongs, further, which set of sample data, characteristics, a model, and a link connected to an external system is used, specifically, a scene corresponding to the target scene information may be acquired, then, the acquired scene may be compared with a pre-stored scene having a set of sample data, characteristics, a model, a link connected to an external system, and the like, and if the scene corresponding to the target scene information is included in a predetermined set of scenes, it may be indicated that the scene corresponding to the target scene information may use a predetermined set of sample data, characteristics, a model, and a link connected to an external system, at this time, a predetermined set of characteristics may be acquired, and a predetermined set of characteristics may be determined based on a predetermined set of characteristics corresponding to a predetermined set of scenes. Meanwhile, a set of sample data corresponding to a preset scene set can be obtained, namely, the corresponding feature types can be based on the preset scene set, corresponding buried points can be set according to the set mode, set positions and the like of the buried points corresponding to the conversion logs and the feature logs determined in the set of sample data corresponding to the preset scene set, then, the set recommended information can be recommended to corresponding users, related behaviors of the users and related features (such as position features and environment features) of the users can be recorded through the buried points, finally, the conversion logs and the feature logs can be generated, and the corresponding conversion logs and the feature logs can be associated based on recorded session identifications, so that sample data can be formed, and the sample data can be the sample data corresponding to the target scene information.

In step S106, feature extraction is performed on the sample data based on the feature class, so as to obtain a target feature corresponding to the sample data.

In implementation, after the sample data corresponding to the target scene information is obtained through the processing in the step S104, feature extraction may be performed on the sample data for different feature classes, for example, for a user feature class, one or more feature types included in the sample data may be preset, and then features of feature types included in the user feature class may be extracted from the sample data. The set of features obtained in the above manner can be used as the target feature corresponding to the sample data. The feature extraction of the sample data may be implemented in various manners, and may be specifically set according to actual situations, which is not limited in the embodiment of the present specification.

In step S108, training the recommendation model corresponding to the scene set by using the target feature, to obtain an information recommendation model corresponding to the target scene information.

The recommended model may be a model constructed based on a certain or multiple preset algorithms, and may be specifically set according to actual situations, which is not limited in the embodiment of the present specification. Correspondingly, the information recommendation model is a model obtained after the recommendation model is trained, so that the information recommendation model is the same as an algorithm used by the recommendation model.

In implementation, based on the above, since the predetermined scene set corresponds to a set of sample data, features, models, and links connected to the external system, the same set of models is used for the scene corresponding to the target scene information and the predetermined scene set, and therefore, the recommendation model corresponding to the predetermined scene set may be trained using the target features extracted in the step S106, so as to obtain a trained recommendation model, and the trained recommendation model may be used as the information recommendation model corresponding to the target scene information.

And then, the accuracy and the like of the information recommendation model corresponding to the target scene information can be verified, and if the verification is passed, the link connected with the external system can be constructed for the information recommendation model corresponding to the target scene information based on the link connected with the external system corresponding to the preset scene set, so that relevant services of information recommendation are provided for corresponding merchants or suppliers and the like. If the verification is not passed, sample data, features and the like can be obtained continuously through the processing procedure, the recommendation model is trained, the trained recommendation model is verified, and the processing procedure is repeated until the trained recommendation model passes the verification.

According to the method for constructing the recommendation model, if the scene corresponding to the target scene information is contained in the preset scene set, sample data corresponding to the target scene information is acquired according to the feature class corresponding to the preset scene set, further target features corresponding to the sample data are obtained, the recommendation model corresponding to the preset scene set is trained by using the target features, and the information recommendation model corresponding to the target scene information is obtained.

Example two

As shown in fig. 3, the embodiment of the present disclosure provides a method for constructing a recommendation model, where an execution body of the method may be a server, where the server may be an independent server, or may be a server cluster formed by a plurality of servers, and the server may be a background server of a website (such as an online shopping website or a shopping application), or may be a server of a service (such as an information recommendation service), or the like. The method can be used for providing a general marketing recommendation model framework so as to improve the online efficiency of the model and other processes. The method specifically comprises the following steps:

In step S302, target scene information for information recommendation is acquired.

In step S304, if the scene corresponding to the target scene information is included in the predetermined scene set, the feature class corresponding to the scene set is acquired.

Wherein the scenes may be divided into different types according to some predetermined manner, one type of scene may include a plurality of specific scenes, for example, the equity information type scene may include a coupon recommendation scene, a prize recommendation scene, etc., the types of scenes are different, and the included scenes may be different. The feature categories corresponding to different types of scenes may be different, in this embodiment, the feature categories may include one or more of a user feature category, a feature category corresponding to recommended information, and a behavior feature category corresponding to recommended information by a user, where the user feature category may be a category corresponding to a user feature, the user feature may mainly include a user portrait feature that tends to be static, and the user portrait feature may be obtained from personal information of the user, for example, may extract corresponding features from information such as gender, age, hobbies, occupation, and the like in the personal information of the user, and in addition, since the user portrait feature is a feature that tends to be static, the user portrait feature may be multiplexed by multiple different types of scenes. The feature class corresponding to the recommended information may be a class corresponding to a feature of the recommended information, the feature of the recommended information may mainly include a basic information class feature of the recommended information and a flow class feature corresponding to the recommended information, the basic information class feature of the recommended information may include a feature corresponding to information such as a name, an information class, a content and the like of the recommended information, the flow class feature corresponding to the recommended information may be a feature for characterizing a degree of interest of the user in the recommended information, for example, a feature corresponding to the recommended information clicked by the user or a feature corresponding to the information not clicked by the user. The behavior characteristics of the user on the recommended information may be mainly related characteristics of the user on the history behavior of the recommended information, for example, characteristics corresponding to the information that the user clicks the recommendation, characteristics corresponding to the information that the user browses the recommendation, and the like.

In step S306, a data embedding point in the recommended information corresponding to the target scene information is determined according to the feature class.

In the implementation, in order to obtain a model corresponding to the target scene information, sample data needs to be built for the model, corresponding data embedded points can be set in recommended information in order to obtain the sample data, operation behaviors of a user on the recommended information can be recorded through the set data embedded points, and in addition, corresponding embedded points can be set in the recommended information based on the determined feature types, so that features corresponding to the feature types can be obtained. Therefore, the data buried point in the recommended information corresponding to the target scene information can be determined based on the determined feature category or the like.

In step S308, sample data corresponding to the target scene information is determined based on the data obtained by the data embedding point.

In the implementation, the recommended information provided with the data embedded point may be sent to the corresponding user in a preset manner, for example, the recommended information provided with the data embedded point may be sent to all registered users, or the recommended information provided with the data embedded point may be sent to the user who has previously accepted the current sample data acquisition, or the recommended information provided with the data embedded point may be sent to the designated user, or the like. Then, whether the data embedded point in the recommended information is triggered can be judged, if a certain data embedded point is triggered, corresponding data can be recorded in a preset log (such as a conversion log or a feature log) and the information recorded in the log can comprise standardized identification of the recommended information, a style (or type, etc.) and a session identification of the recommended information, meanwhile, the corresponding session identification is required to be recorded in the feature log, the corresponding conversion log and the feature log can be associated based on the recorded session identification, so that sample data can be formed, and finally the obtained sample data can be used as sample data corresponding to the target scene information.

In step S310, feature extraction is performed on the sample data based on the feature class, so as to obtain a target feature corresponding to the sample data.

In step S312, according to the feature type included in the target feature, a corresponding feature extraction process is performed on the target feature, so as to obtain an extracted feature.

In practice, in order to secure the stability of the trained model, and considering that if the recommended model is a deep learning model, character-type features will not be directly learned and higher-order features will not be sufficiently extracted, it is necessary to perform feature ID processing, feature derivation processing, and the like on target features extracted from sample data to perform feature extraction processing, so that features for training the model are more discretized.

The specific processing of step S312 may be varied, and an optional processing manner is provided below, and in general, the commonly used feature extraction processing is implemented through manual experience, however, the feature extraction processing performed through the manual experience may make the feature extraction inefficient, so that the feature extraction process may be simplified appropriately, and corresponding feature operators may be adapted for different feature types based on whole-table statistics and feature binning, which may specifically include the following cases one to three.

In the first case, if the target feature includes a character type feature, discretizing the character type feature in the target feature to obtain a first feature, and taking the first feature as an extracted feature.

The character type feature may be a related feature in which the character is contained.

In implementation, considering that the characters are related and continuous, in order to ensure the stability of the training model, discretizing the character type features in the target features, extracting the features after discretizing to obtain first features, and taking the first features as extracted features.

And secondly, if the target feature comprises the numerical type feature, carrying out box division processing and normalization processing on the numerical type feature in the target feature to obtain a second feature, and taking the second feature as the extracted feature.

The numerical feature may be a related feature including a numerical value therein.

In implementation, if the target feature includes a numerical feature, the numerical feature in the target feature is subjected to binning, that is, discrete binning or segmentation can be performed on the numerical feature in the target feature, where the binning method may include multiple methods, for example, a supervised binning method, an unsupervised binning method, and the like, where the supervised binning method may further include a chi-square binning method, a minimum entropy method, and the like, and the unsupervised binning method may include an equidistant binning method, an equal frequency sharing method, and the like, and specifically, the numerical feature in the target feature may be subjected to binning by selecting a corresponding binning method according to an actual situation. In addition, the numerical type feature in the target feature can be normalized, and finally a second feature can be obtained and used as an extracted feature.

And thirdly, if the target feature comprises the predefined feature, carrying out Cartesian product cross extraction processing on the predefined feature in the target feature to obtain a third feature, and taking the third feature as an extracted feature.

The predefined features may be a certain organization structure or a feature set by an individual according to actual needs, etc.

It should be noted that, the automatic generation of the feature extraction configuration can be realized through the SQL statement, so that the efficiency of feature extraction can be greatly improved.

In step S314, training is performed on the recommendation model corresponding to the predetermined scene set by using the extracted features, so as to obtain an information recommendation model corresponding to the target scene information.

The recommendation model may be a Deep learning model, and the recommendation model may specifically be a Wide & Deep model, a Deep fm (Deep Factorization Machine, depth factorizer) model, an AFM (Attentional Factorization Machine, attention factorizer) model, a DIN (Deep Interest Network, depth interest network) model, or the like, and the corresponding recommendation model may specifically be selected according to actual situations.

It should be noted that, in the process of training the recommended model corresponding to the predetermined scene set by using the extracted features or in the process of training the recommended model corresponding to the target scene information by using the target features, the model may be debugged and/or trained by using a single-machine-node debugging mechanism to debug the recommended model, so that the network structure of the recommended model may be debugged by using the interactive advantage of the jupyter-node, and meanwhile, the recommended model may be trained by using a predetermined yarn distributed training mechanism, so that the training of the big data set may be realized by using the computing capability of the distributed cluster. In practical application, the packaging of tensorflow estimator (or the development of a neural network based on keras, etc.) can be realized, and a single device and a distributed system can use the same set of processing codes and simultaneously package the data processing and model evaluation modules, so that an algorithm engineer only needs to pay attention to the core network structure of the recommendation model, and the development complexity is greatly simplified.

And then, according to a set of links corresponding to the preset scene set and connected with an external system, the external system is uniformly connected through an operation platform of the recommendation model, so that the serial connection and the closed loop of the links are finished, and an algorithm engineer is more concerned with the recommendation flow.

Deployment of the information recommendation model may be deployed via a model-based predictive platform (e.g., c++ compiled version, etc.). In order to improve the deployment efficiency, the model development platform and the model deployment platform can be communicated, and the recommendation model and the feature configuration generated by the model development platform can be conveniently used for deploying the information recommendation model. Meanwhile, the efficiency of the model deployment test link is improved conveniently through the custom SQL component mock test data on the model development platform. The information recommendation is mainly realized on a recommendation platform, and the recommendation platform can integrate core functions such as feature view configuration, recall configuration, model call, feature query, feature log backflow, AB experiment and the like, and is used for arranging an information recommendation flow based on a preset information recommendation framework, so that a complete recommendation flow can be built through a small number of codes. The feature view can support a set of online features shared by a plurality of scenes, so that the reusability is greatly enhanced. And packaging and optimizing the model call, so that a small amount of codes can realize the model call, and parallel optimization can be performed on performance. The feature log plug-in is introduced, so that the feature log can flow back more smoothly, and the expansion of additional features can be supported. In the aspect of AB experiments, recommended scenes can be automatically associated, experiment control is carried out through configuration fields defined in an experiment scheme, and the whole-flow AB experiments (including recall, sequencing, rearrangement and the like) are supported. After the whole link is built, only the AB experimental configuration needs to be modified in the online iteration process of the model, so that the high efficiency and convenience are achieved. The recall link can introduce search engine capability in the content ordering item, build a service system- > search engine-recall platform- > recommendation platform- > link of the service system, and reserve the capability of multipath real-time recall.

According to the method for constructing the recommendation model, if the scene corresponding to the target scene information is contained in the preset scene set, sample data corresponding to the target scene information is acquired according to the feature class corresponding to the preset scene set, further target features corresponding to the sample data are obtained, the recommendation model corresponding to the preset scene set is trained by using the target features, and the information recommendation model corresponding to the target scene information is obtained. In addition, in practical application, the packaging such as tensorflow estimator (or developing a neural network based on the keras, etc.) can be realized, and a single device and a distributed system can use the same set of processing codes and package the data processing and model evaluation modules at the same time, so that an algorithm engineer only needs to pay attention to the core network structure of the recommendation model, and the development complexity is greatly simplified. And then, according to a set of links corresponding to the preset scene set and connected with an external system, the external system is uniformly connected through an operation platform of the recommendation model, so that the serial connection and the closed loop of the links are finished, and an algorithm engineer is more concerned with the recommendation flow.

Example III

The above method for constructing a recommendation model provided in the embodiment of the present disclosure further provides a device for constructing a recommendation model based on the same concept, as shown in fig. 4.

The recommendation model constructing device comprises: a scene information acquisition module 401, a sample acquisition module 402, a feature extraction module 403, and a training module 404, wherein:

a scene information acquisition module 401, configured to acquire target scene information for information recommendation;

a sample acquiring module 402, configured to acquire a feature class corresponding to a scene set if a scene corresponding to the target scene information is included in a predetermined scene set, and acquire sample data corresponding to the target scene information according to the feature class;

the feature extraction module 403 is configured to perform feature extraction on the sample data based on the feature class, so as to obtain a target feature corresponding to the sample data;

and the training module 404 is configured to train the recommendation model corresponding to the scene set by using the target feature, so as to obtain an information recommendation model corresponding to the target scene information.

In the embodiment of the present disclosure, the feature class includes one or more of a user feature class, a feature class corresponding to recommended information, and a behavior feature class of the user on the recommended information.

In the embodiment of the present disclosure, the sample acquiring module 402 includes:

In an embodiment of the present disclosure, the apparatus further includes:

In an embodiment of the present disclosure, the feature extraction module includes:

In the embodiment of the present specification, the recommendation model is a deep learning model.

In the embodiment of the present specification, the recommended model is a Wide & Deep model, a Deep fm model of a depth factorer, an AFM model of an attention factorer, or a DIN model of a depth interest network.

In an embodiment of the present disclosure, the apparatus further includes:

the model debugging module is used for debugging the recommended model based on a preset single-machine NoteBook debugging mechanism; and/or the number of the groups of groups,

and the distributed training module is used for training the recommendation model based on a preset Yarn distributed training mechanism.

According to the embodiment of the specification, through obtaining the target scene information for information recommendation, if the scene corresponding to the target scene information is contained in the preset scene set, sample data corresponding to the target scene information is obtained according to the feature class corresponding to the preset scene set, further target features corresponding to the sample data are obtained, the recommendation model corresponding to the preset scene set is trained by using the target features, and the information recommendation model corresponding to the target scene information is obtained, so that the feature, the sample data and the recommendation model corresponding to the preset scene set are standardized, a general marketing recommendation model frame is provided for each scene (the scene corresponding to the target scene information) in the preset scene set, and the construction mode and the construction flow of the same set of recommendation model can be multiplexed for each scene in the preset scene set, so that the construction flow of the recommendation model is simplified, and the online efficiency of the model is greatly saved. In addition, in practical application, the packaging such as tensorflow estimator (or developing a neural network based on the keras, etc.) can be realized, and a single device and a distributed system can use the same set of processing codes and package the data processing and model evaluation modules at the same time, so that an algorithm engineer only needs to pay attention to the core network structure of the recommendation model, and the development complexity is greatly simplified. And then, according to a set of links corresponding to the preset scene set and connected with an external system, the external system is uniformly connected through an operation platform of the recommendation model, so that the serial connection and the closed loop of the links are finished, and an algorithm engineer is more concerned with the recommendation flow.

Example IV

The above device for constructing a recommendation model provided in the embodiment of the present disclosure further provides a device for constructing a recommendation model based on the same concept, as shown in fig. 5.

The recommendation model construction device may be a server provided in the above embodiment.

The recommendation model may be configured or configured to vary significantly, and may include one or more processors 501 and memory 502, where the memory 502 may store one or more stored applications or data. Wherein the memory 502 may be transient storage or persistent storage. The application program stored in the memory 502 may include one or more modules (not shown in the figures), each of which may include a series of computer-executable instructions in the construction device for the recommendation model. Still further, the processor 501 may be configured to communicate with the memory 502 and execute a series of computer executable instructions in the memory 502 on a build device of the recommendation model. The recommendation model building device may also include one or more power supplies 503, one or more wired or wireless network interfaces 504, one or more input/output interfaces 505, and one or more keyboards 506.

In particular, in this embodiment, the recommendation model building apparatus includes a memory, and one or more programs, where the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the recommendation model building apparatus, and configured to be executed by the one or more processors, the one or more programs including computer executable instructions for:

acquiring target scene information for information recommendation;

In this embodiment of the present disclosure, the obtaining, according to the feature class, sample data corresponding to the target scene information includes:

In this embodiment of the present specification, further includes:

In this embodiment of the present disclosure, the performing, according to the feature type included in the target feature, a corresponding feature extraction process on the target feature to obtain an extracted feature includes:

In this embodiment of the present disclosure, in the training the recommendation model corresponding to the target scene information using the target feature, the method further includes:

The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

In the 90 s of the 20 th century, improvements to one technology could clearly be distinguished as improvements in hardware (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software (improvements to the process flow). However, with the development of technology, many improvements of the current method flows can be regarded as direct improvements of hardware circuit structures. Designers almost always obtain corresponding hardware circuit structures by programming improved method flows into hardware circuits. Therefore, an improvement of a method flow cannot be said to be realized by a hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (e.g., field programmable gate array (Field Programmable Gate Array, FPGA)) is an integrated circuit whose logic function is determined by the programming of the device by a user. A designer programs to "integrate" a digital system onto a PLD without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, such programming is mostly implemented by using "logic compiler" software, which is similar to the software compiler used in program development and writing, and the original code before the compiling is also written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), but not just one of the hdds, but a plurality of kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), lava, lola, myHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog are currently most commonly used. It will also be apparent to those skilled in the art that a hardware circuit implementing the logic method flow can be readily obtained by merely slightly programming the method flow into an integrated circuit using several of the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable logic controllers, and embedded microcontrollers, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller may thus be regarded as a kind of hardware component, and means for performing various functions included therein may also be regarded as structures within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.

The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functionality of the units may be implemented in one or more software and/or hardware when implementing one or more embodiments of the present description.

It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, one or more embodiments of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present description are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.

One or more embodiments of the present specification may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. One or more embodiments of the present description may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments.

The foregoing is merely exemplary of the present disclosure and is not intended to limit the present disclosure. Various modifications and alterations to this specification will become apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, or the like, which are within the spirit and principles of the present description, are intended to be included within the scope of the claims of this document.

Claims

1. A method of constructing a recommendation model, the method comprising:

acquiring target scene information for information recommendation;

training a recommendation model corresponding to the scene set by using the target features to obtain an information recommendation model corresponding to the target scene information;

the method comprises the steps of providing a general recommendation model frame corresponding to the preset scene set in advance, wherein the general recommendation model frame corresponding to the scene set comprises: the feature category corresponding to the scene set and the recommendation model corresponding to the scene set;

the feature categories corresponding to the scene set comprise: user feature class, feature class corresponding to recommended information, and behavior feature class of the user on the recommended information;

the obtaining sample data corresponding to the target scene information according to the feature class includes:

Determining a setting mode and a setting position of a data embedding point of recommended information corresponding to the target scene information according to the characteristic category;

setting data embedded points in the recommended information according to the setting mode and the setting positions of the data embedded points, sending the recommended information after the data embedded points are set to corresponding users, and taking the data obtained based on the data embedded points as sample data corresponding to the target scene information; wherein, the data obtained based on the data embedded point comprises: and the characteristics corresponding to the characteristic categories.

2. The method of claim 1, the method further comprising:

3. The method according to claim 2, wherein the performing, according to the feature type included in the target feature, a corresponding feature extraction process on the target feature to obtain an extracted feature includes:

4. A method according to any one of claims 1-3, the recommendation model being a deep learning model.

5. The method of claim 4, the recommended model is a Wide & Deep model, a depth factorizer Deep fm model, an attention factorizer AFM model, or a depth interest network DIN model.

6. The method of claim 5, wherein in the training the recommendation model corresponding to the target scene information using the target feature, the method further comprises:

7. A recommendation model building apparatus, the apparatus comprising:

the training module is used for training the recommendation model corresponding to the scene set by using the target features to obtain an information recommendation model corresponding to the target scene information;

8. The apparatus of claim 7, the apparatus further comprising:

9. The apparatus of claim 8, the feature extraction module comprising:

10. The apparatus of any of claims 7-9, the recommendation model being a deep learning model.

11. A recommendation model construction apparatus, the recommendation model construction apparatus comprising:

a processor; and

Acquiring target scene information for information recommendation;