CN112379919B

CN112379919B - Service customization method and device, electronic equipment and storage medium

Info

Publication number: CN112379919B
Application number: CN202011320234.6A
Authority: CN
Inventors: 刘亚虹
Original assignee: Individual
Current assignee: Individual
Priority date: 2020-11-23
Filing date: 2020-11-23
Publication date: 2024-04-09
Anticipated expiration: 2040-11-23
Also published as: CN112379919A

Abstract

The application discloses a service customization method, a device, electronic equipment and a storage medium, which belong to the technical field of Internet, and the method comprises the following steps: receiving a service customization request, wherein the request comprises domain description information of a target platform, service description information of customized services and a data sample in the target platform, determining a target domain to which the target platform belongs according to the domain description information, selecting a domain service model for providing customized services corresponding to the service description information from the target domain, performing migration learning on the service relationship of customized services in the target platform by utilizing the service relationship and the data sample of customized services in the target domain learned by the domain service model, obtaining a special service model of the target platform, and providing customized services by utilizing the special service model, wherein the domain service model is obtained by performing migration learning on the service relationship of customized services in different domains including the target domain learned by utilizing a general service model.

Description

Service customization method and device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of internet technologies, and in particular, to a service customization method, device, electronic device, and storage medium.

Background

With the rapid development of the internet industry, almost all enterprises develop online business operation channels. Under the condition that a plurality of enterprises develop online business operation channels, how to improve the online operation effect is a problem that each enterprise needs to consider.

Generally, each enterprise has its own online service platform, and the operation effect of one platform has a direct relationship with the service accuracy of the platform, such as recommendation accuracy and search accuracy, so that a better online operation effect is to realize more accurate recommendation service and more accurate search service. However, for small and medium enterprises, they may not have their own technical team or only a less experienced technical team, and it is difficult for them to promote their own platform's recommended service effects and/or search service effects in a short time.

Therefore, how to reduce the development threshold and enable middle and small enterprises to improve the service accuracy of the platform in a short time becomes a technical problem to be solved.

Disclosure of Invention

The embodiment of the application provides a service customization method, a device, electronic equipment and a storage medium, which are used for solving the technical problem that the service accuracy of a platform is difficult to improve in a short time in the related technology.

In a first aspect, an embodiment of the present application provides a service customization method, including:

receiving a service customization request, wherein the service customization request at least comprises field description information of a target platform, service description information of customized service and a data sample in the target platform required by the customized service;

determining the target domain to which the target platform belongs according to the domain description information;

selecting a domain service model for providing customized services corresponding to the service description information from the target domain, wherein the domain service model is obtained by performing migration learning on the service relationship of the customized services in the target domain by utilizing the service relationship of the customized services in different domains including the target domain learned by a general service model, and the service relationship of the customized services is a feature matching relationship between service input data and service response data of a specified uniform service;

performing migration learning on the service relationship of the customized service in the target platform by using the service relationship of the customized service in the target domain and the data sample learned by the domain service model to obtain a special service model of the target platform;

And providing customized services in the target platform by using the special service model.

In one possible implementation manner, using the service relationship of the customized service in the target domain learned by the domain service model and the data sample, performing migration learning on the service relationship of the customized service in the target platform to obtain a dedicated service model of the target platform, where the migration learning includes:

screening candidate service response data from a service response database of the target platform by utilizing the service relation of the customized service in the target domain and the service input data sample of the customized service in the target platform, which are learned by the domain service model;

sorting the candidate service response data;

and adjusting model parameters of the domain service model according to the sorting result and preset service response data corresponding to the service input data sample until the service accuracy of the domain service model is determined to reach a set value, and determining the domain service model as a special service model of the target platform.

In one possible implementation manner, if the service customization request further includes an optimization target of the customized service, ranking each candidate service response data includes:

Calculating the matching degree between each candidate service response data and the optimization target of the customized service;

and sequencing the candidate service response data according to the matching degree between the candidate service response data and the optimization target of the customized service.

In one possible implementation, calculating the matching degree between each candidate service response data and the optimization target of the customized service includes:

when the customized service is a search service, analyzing historical user operation data corresponding to each candidate service response data, determining a first matching degree between the candidate service response data and the optimization target, comparing the candidate service response data with a user interest characteristic sample corresponding to the service input data sample, and determining a second matching degree between the candidate service response data and the optimization target; determining the matching degree between the candidate service response data and the optimization target according to the first matching degree and the second matching degree;

when the customized service is recommended service, analyzing historical user operation data corresponding to each candidate service response data, determining a first matching degree between the candidate service response data and the optimization target, and determining the matching degree between the candidate service response data and the optimization target according to the first matching degree.

In one possible embodiment, when the customized service is a recommended service or a search service, the service response data of the customized service is a multimedia content feature, and further comprising:

selecting a domain content understanding model for providing content understanding services from the target domain, wherein the domain content understanding model is obtained by performing migration learning on service relationships of the content understanding services in the target domain by utilizing service relationships of the content understanding services in different domains including the target domain learned by a general content understanding model, and the service relationships of the content understanding services refer to content understanding relationships between multimedia data and multimedia content characteristics; and

the multimedia content features in the multimedia feature library of the target platform are obtained by performing content understanding on the multimedia data in the target platform by using a special content understanding model, wherein the special content understanding model is obtained by performing migration learning on the service relationship of the content understanding service in the target platform by using the service relationship of the content understanding service in the target domain learned by the domain content understanding model.

In one possible implementation, when the customized service is a recommended service, the service input data of the customized service is a user interest feature; when the customized service is a search service, the service input data of the customized service is a search word feature sample, the search word feature sample corresponding to a user interest feature, and further comprising:

Selecting a domain user interest model for providing user interest analysis services from the target domain, wherein the domain user interest model is obtained by performing migration learning on the service relationship of the user interest analysis services in the target domain by utilizing the service relationship of the user interest analysis services in different domains including the target domain learned by a general user interest model, and the service relationship of the user interest analysis services refers to the association relationship between user interest characterization data and user interest characteristics; and

the user interest feature samples of the recommended service in the target platform and/or the user interest feature samples corresponding to the search term feature samples of the search service in the target platform are obtained by utilizing a special user interest model to conduct interest analysis on the user interest characterization data in the target platform, wherein the special user interest model is obtained by utilizing the service relationship of the user interest analysis service in the target field learned by the field user interest model to conduct migration learning on the service relationship of the user interest analysis service in the target platform.

In one possible implementation manner, after providing the customized service in the target platform by using the dedicated service model, the method further includes:

Acquiring new user interest characterization data in the target platform;

and updating the special service model by utilizing the new user interest characterization data.

In a second aspect, an embodiment of the present application provides a service customization apparatus, including:

the system comprises a receiving module, a service customization request module and a storage module, wherein the receiving module is used for receiving a service customization request, and the service customization request at least comprises field description information of a target platform, service description information of customized service and a data sample in the target platform required by the customized service;

the determining module is used for determining the target domain to which the target platform belongs according to the domain description information;

a selection module, configured to select, from the target domain, a domain service model that provides a customized service corresponding to the service description information, where the domain service model is obtained by performing migration learning on a service relationship of the customized service in the target domain by using a service relationship of the customized service in different domains including the target domain learned by a generic service model, where the service relationship of the customized service is a feature matching relationship between service input data and service response data of a specific uniform service;

the training module is used for performing migration learning on the service relationship of the customized service in the target platform by utilizing the service relationship of the customized service in the target field and the data sample learned by the field service model to obtain a special service model of the target platform;

And the service module is used for providing customized service in the target platform by utilizing the special service model.

In one possible implementation, the training module is specifically configured to:

sorting the candidate service response data;

In one possible implementation manner, if the service customization request further includes an optimization target of the customized service, the training module is specifically configured to:

In one possible implementation, when the customized service is a recommended service or a search service, the service response data of the customized service is a multimedia content feature;

The selection module is further configured to select a domain content understanding model for providing a content understanding service from the target domain, where the domain content understanding model is obtained by performing migration learning on a service relationship of the content understanding service in the target domain by using a service relationship of the content understanding service in different domains including the target domain learned by using a general content understanding model, where the service relationship of the content understanding service refers to a content understanding relationship between multimedia data and multimedia content features; and

In one possible implementation, when the customized service is a recommended service, the service input data of the customized service is a user interest feature; when the customized service is a search service, service input data of the customized service is a search word feature sample, and the search word feature sample corresponds to a user interest feature;

The selection module is further configured to select a domain user interest model that provides a user interest analysis service from the target domain, where the domain user interest model is obtained by performing migration learning on a service relationship of the user interest analysis service in the target domain by using a service relationship of the user interest analysis service in different domains including the target domain learned by using a general user interest model, where the service relationship of the user interest analysis service refers to an association relationship between user interest characterization data and user interest features; and

In one possible embodiment, the method further comprises:

The feedback module is used for acquiring new user interest characterization data in the target platform after providing customized service in the target platform by utilizing the special service model; and updating the special service model by utilizing the new user interest characterization data.

In a third aspect, an embodiment of the present application provides an electronic device, including: at least one processor, and a memory communicatively coupled to the at least one processor, wherein:

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the service customization method described above.

In a fourth aspect, embodiments of the present application provide a storage medium, which when executed by a processor of an electronic device, is capable of performing the above-described service customization method.

In the embodiment of the application, the service relationship of the customized service in different fields including the target field is learned in advance to obtain a general service model, and then the service relationship of the customized service in the target field is migrated and learned by utilizing the service relationship of the customized service in the different fields learned by the general service model to obtain a field service model, wherein the service relationship of the customized service is a characteristic matching relationship between service input data and service response data of the appointed uniform service. Subsequently, when a service customization request of any target platform is received, determining a target domain to which the target platform belongs according to domain description information contained in the service customization request, selecting a domain service model for providing customization service corresponding to the service description information contained in the service customization request from the target domain, and performing migration learning on the service relationship of the customization service in the target platform by utilizing a service relationship of the customization service in the target domain learned by the domain service model and a data sample in the target platform required by the customization service contained in the service customization request to obtain a special service model of the target platform, so that the customization service is provided in the target platform by utilizing the special service model.

Therefore, a developer can customize the service for the target platform quickly by only providing the domain description information of the target platform, the service description information of the customized service and the data sample in the target platform required by the customized service, without concerning the bottom implementation of the customized service, and the development threshold of the service is lower. The general service model learns the service relation of the customized service in different fields, the field service model continuously learns the service relation of the customized service in the target field on the basis of the general service model, the special service model continuously learns the service relation of the customized service in the target platform belonging to the target field on the basis of the field service model, the process of gradually learning the service relation of the customized service from easy to difficult and from thick to thin is thinned, the learning targets of the service models of all stages are thinned, the convergence speed of the service models of all stages can be accelerated, and the machine learning rule is also more met, so that the convergence speed and the service accuracy of the finally obtained special service model can be improved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application. In the drawings:

Fig. 1 is a flowchart of a service customization method according to an embodiment of the present application;

FIG. 2 is a flowchart of a method for training to obtain a dedicated service model according to an embodiment of the present application;

FIG. 3 is a flowchart of another method for training to obtain a dedicated service model according to an embodiment of the present application;

FIG. 4 is a building block diagram of a recommendation service and a search service according to an embodiment of the present application;

FIG. 5 is a schematic diagram of a service construction process according to an embodiment of the present application;

FIG. 6 is a schematic diagram of a customization process for a recommendation service provided by an embodiment of the present application;

FIG. 7 is a schematic diagram of a customization process of a search service provided by an embodiment of the present application;

fig. 8 is a schematic structural diagram of a service customization device according to an embodiment of the present application;

fig. 9 is a schematic hardware structure of an electronic device for implementing a service customization method according to an embodiment of the present application.

Detailed Description

In order to solve the technical problem that in the related art, the service accuracy of a platform is difficult to improve in a short time, the embodiment of the application provides a service customization method, a device, electronic equipment and a storage medium.

The preferred embodiments of the present application will be described below with reference to the accompanying drawings of the specification, it being understood that the preferred embodiments described herein are for illustration and explanation only, and are not intended to limit the present application, and embodiments and features of embodiments of the present application may be combined with each other without conflict.

In order to facilitate understanding of the present application, the present application refers to the technical terms:

the service relation of the customized service is a characteristic matching relation between service input data and service response data of the appointed uniform service. Taking the example that the customized service is a recommended service, the service input data of the recommended service is a user interest feature and the service response data is a multimedia content feature, so that the service relationship of the recommended service refers to a feature matching relationship between the user interest feature and the multimedia content feature of the recommended service. Taking the customized service as a search service as an example, the service input data of the search service is a search word feature, the service response data is a multimedia content feature, so the service relationship of the search service is a feature matching relationship between the search word feature and the multimedia content feature of the search service.

Multimedia data generally refers to data in various media forms including text, sound, and images, such as live video, an album, short video, and the like, which are multimedia data.

Multimedia content features, the data obtained after content understanding of multimedia data is called multimedia content features.

Fig. 1 is a flowchart of a service customization method according to an embodiment of the present application, including the following steps:

S101: and receiving a service customization request, wherein the service customization request at least comprises domain description information of a target platform, service description information of a customization service and a data sample in the target platform required by the customization service.

In particular, the domain description information may be a domain name such as an education domain, an automobile domain, or the like, or may be a domain identifier such as domain1 (indicating an education domain) or domain2 (indicating an automobile domain). Similarly, the service customization information may be a service name such as a recommended service, a search service, or the like, or may be a service identification such as a service1 (indicating a recommended service), a service2 (indicating a search service), or the like. Of course, the domain description information and the service customization information may also be other manifestations, which are not limited herein.

In practical applications, a user may customize one service, such as a recommendation service or a search service, or may customize a plurality of services, such as a recommendation service and a search service, at a time, so that the service customization information may be service customization information of one service or service customization information of a plurality of services. The embodiments of the present application are not limited in this regard.

S102: and determining the target domain to which the target platform belongs according to the domain description information.

In practical applications, each domain has a domain service model for providing customized services, and the domain service models for providing customized services in different domains are different. Taking the customized service as a recommended service as an example, a domain service model for providing the recommended service in the automobile domain is mainly used for recommending videos related to automobiles, and a domain service model for providing the recommended service in the education domain is mainly used for recommending videos related to education, and the two domain service models for providing the recommended service in the two domains are different due to the fact that the video content in the two domains is large in difference.

In order to better provide the customized service for the target platform, the target domain to which the target platform belongs can be determined according to the domain description information, and then a domain service model for providing the customized service is selected from the target domain.

S103: selecting a domain service model for providing customized services corresponding to the service description information from the target domain, wherein the domain service model is obtained by utilizing the service relationship of the customized services in different domains including the target domain learned by the general service model to perform migration learning on the service relationship of the customized services in the target domain, and the service relationship of the customized services is a feature matching relationship between service input data and service response data of the appointed uniform service.

Taking the customized service as a recommended service as an example, selecting a domain service model corresponding to the customized service, namely selecting a domain service model providing the recommended service, subsequently, the domain service model providing the recommended service is called a domain recommendation model, a general service model for training the domain recommendation model is called a general recommendation model, and a special service model trained by using the domain recommendation model is called a special recommendation model.

Taking the customized service as a search service as an example, selecting a domain service model corresponding to the customized service, that is, selecting a domain service model providing the search service, subsequently, the domain service model providing the search service is referred to as a domain search model, a general service model for training the domain search model is referred to as a general search model, and a special search model trained by using the domain search model is referred to as a special search model.

S104: and performing migration learning on the service relationship of the customized service in the target platform by utilizing the service relationship of the customized service in the target domain learned by the domain service model and the data sample in the target platform required by the customized service to obtain a special service model of the target platform.

In specific implementation, the specific service model may be trained according to the flow shown in fig. 2, where the flow includes the following steps:

S201a: and screening candidate service response data from the service response database of the target platform by utilizing the service relation of the customized service in the target domain and the service input data sample of the customized service in the target platform, which are learned by the domain service model.

S202a: and sequencing the candidate service response data.

In specific implementation, the candidate service response data may be ranked according to a preset rule, and then the candidate service response data may be ranked according to a sequence from high to low in matching degree, where the preset rule includes heat, release time, and matching degree between service input data.

S203a: and adjusting model parameters of the domain service model according to the sequencing result and preset service response data corresponding to the service input data sample until the service accuracy of the domain service model reaches a set value, and determining the domain service model as a special service model of the target platform.

In addition, in order to provide deep service customization meeting the service requirement for the target platform, the service customization request may further include an optimization target of the customized service, and at this time, a dedicated service model may be trained according to a process shown in fig. 3, where the process includes the following steps:

S301a: and screening candidate service response data from the service response database of the target platform by utilizing the service relation of the customized service in the target domain and the service input data sample of the customized service in the target platform, which are learned by the domain service model.

S302a: a degree of matching between each candidate service response data and an optimization objective of the customized service is calculated.

In practical applications, the customized service may have more than one optimization objective, and the optimization objectives for the customized service may be different in different fields. For example, the optimization goal of the recommended service in the automotive field is the total amount of automobile deals; and the optimal goal of the recommendation service in the education field is the completion rate of the education video. For another example, the optimization targets of the search service in the automobile field are the browsing duration of automobile videos and the total amount of automobile deals; while the search service is optimized in the educational field with the goal of relevance and the rate of completion of the educational video.

When the customized service is a recommended service, historical user operation data corresponding to each candidate service response data can be analyzed, a first matching degree between the candidate service response data and an optimization target of the recommended service is determined, and then the matching degree between the candidate service response data and the optimization target of the recommended service is determined according to the first matching degree. For example, the matching degree between the candidate service response data and the optimization target of the recommended service=β×the first matching degree, where β is a constant predetermined by the technician.

Taking the example that the optimization objective of the recommended service in the automobile field is the total amount of automobile traffic, for each candidate service response data (multimedia data related to automobiles), the ordering probability (namely, the first matching degree) after the user views the candidate service response data can be calculated according to the historical user operation data such as the historical viewing times and the historical ordering times corresponding to the candidate service response data, and then the ordering probability is determined as the matching degree between the candidate service response data and the user interest feature sample, namely, the first matching degree is directly used as the matching degree between the candidate service response data and the optimization objective of the recommended service.

In this way, after candidate service response data conforming to the user interest feature samples are screened out, the matching degree between the candidate service response data and the optimization target of the recommended service is determined by combining historical user operation data such as purchase condition, heat and the like of each candidate service response data, and the candidate service response data are displayed according to the sequence from high to low in matching degree, so that the optimization target of the recommended service is facilitated to be realized.

When the customized service is a search service, historical user operation data corresponding to each candidate service response data can be analyzed, first matching degree between the candidate service response data and an optimization target of the search service is determined, user interest feature samples corresponding to service input data samples (search word feature samples at this time) are compared, second matching degree between the candidate service response data and the optimization target of the search service is determined, and matching degree between the candidate service response data and the optimization target of the search service is determined according to the first matching degree and the second matching degree. For example, the matching degree between the candidate service response data and the optimization target of the search service=α+the first matching degree + (1- α) ×the second matching degree, where α is a decimal predetermined by the technician.

Taking relevance and the completion rate of education videos as an example of the optimization targets of the search service in the education field, for each candidate service response data (multimedia data related to teaching), the relevance degree between the candidate service response data and the characteristic sample of the search word can be calculated to be used as a first matching degree, the completion rate of the current user watching the candidate service response data is calculated to be used as a second matching degree according to the historical user operation data corresponding to the candidate service response data, such as a historical user watching record, and then the matching degree between the candidate service response data and the optimization targets of the search service is calculated according to the first matching degree and the second matching degree.

In this way, after candidate service response data conforming to the search term feature samples are screened out, the matching degree between the candidate service response data and the optimization target of the search service is determined by combining historical user operation data such as purchase condition, heat and corresponding user interest feature samples of each candidate service response data, and the candidate service response data are displayed according to the sequence from high to low in matching degree, so that the optimization target of the search service is facilitated to be realized. And, can show different candidate service response data ordering conditions for different users searching the same search word, utilize further approaching to promote the optimization goal of searching the service.

S303a: and sequencing the candidate service response data according to the matching degree between the candidate service response data and the optimization target of the customized service.

In general, the candidate service response data may be ordered in order of high-to-low matching degree to achieve the objective of optimizing the customized service.

S304a: and adjusting model parameters of the domain service model according to the sequencing result and preset service response data corresponding to the service input data sample until the service accuracy of the domain service model reaches a set value, and determining the domain service model as a special service model of the target platform.

In the embodiment of the application, the domain service model learns the service relationship of the customized service in the target domain, the target platform is a finer branch in the target domain, the service relationship in the target platform is finer and more diversified than the service relationship in the target domain, and after the relevant data of the service relationship in the target platform are added, the service relationship of the customized service in the target domain learned by the domain service model is thinned and corrected by using the unique service data of the target platform, so that the special service model of the target platform can be quickly established on the basis of the domain service model, and the service accuracy of the special service model can be ensured.

S105: custom services are provided in the target platform using a dedicated service model.

In implementation, model parameters of the special service model can be deployed on public cloud or private cloud of the target platform, so that customized service is provided for the target platform by using the special service model.

In practical application, the service response data of the recommendation service or the search service is a multimedia content feature, and when the recommendation service or the search service is customized for the target platform, the multimedia content feature of the multimedia data in the target platform needs to be recommended or searched, so that the content of the multimedia data in the target platform can be understood in advance, and the multimedia content feature of the target platform can be obtained.

In the implementation, the multimedia content features of the target platform can be obtained by a non-transfer learning mode or a transfer learning mode.

When the multimedia content features of the target platform are obtained by using the migration learning method, in S103, a domain content understanding model for providing the content understanding service may be selected from the target domain, where the domain content understanding model is obtained by performing migration learning on the service relationships of the content understanding service in the target domain by using the service relationships of the content understanding service in different domains including the target domain learned by the general content understanding model, where the service relationships of the content understanding service refer to the content understanding relationships between the multimedia data and the multimedia content features.

And then, performing migration learning on the service relationship of the content understanding service in the target platform by utilizing the service relationship of the content understanding service in the target domain and the content understanding sample in the target platform learned by the domain content understanding model, so as to obtain a special content understanding model of the target platform, and performing content understanding on each piece of multimedia data in the target platform by utilizing the special content understanding model, so as to obtain the multimedia content characteristics of the target platform, wherein the content understanding sample comprises the multimedia data and preset multimedia characteristics corresponding to the multimedia data.

In addition, when the customized service is a recommended service, service input data of the customized service is a user interest feature; when the customized service is a search service, the service input data of the customized service is a search word feature sample, and the search word feature sample also corresponds to the user interest feature. That is, when customizing the recommendation service or the search service for the target platform, the user interest feature sample in the target platform is used, so that interest analysis can be performed on the user interest feature data in the target platform in advance to obtain the user interest feature in the target platform, and then the user interest feature sample is selected from the user interest feature, where the user interest feature data includes static portrait data (such as age, sex, geographical location, etc.) and dynamic portrait data (such as which videos are browsed, which commodities are reviewed, which commodities are purchased, and video browsing duration, etc.) of the user.

In specific implementation, the user interest feature in the target platform can be obtained by using a non-transfer learning mode or a transfer learning mode.

When the user interest feature in the target platform is obtained by using the migration learning manner, in S103, a domain user interest model for providing the user interest analysis service may be selected from the target domain, where the domain user interest model is obtained by using the service relationship of the user interest analysis service in different domains including the target domain learned by the general user interest model to migrate and learn the service relationship of the user interest analysis service in the target domain, where the service relationship of the user interest analysis service refers to the association relationship between the user interest characterization data and the user interest feature.

And then, performing migration learning on the service relationship of the user interest analysis service in the target platform by utilizing the service relationship of the user interest analysis service in the target domain learned by the domain user interest model and the user interest analysis sample in the target platform, so as to obtain the special user interest model. And then, carrying out interest analysis on the user interest characterization data of each user in the target platform by using a special user interest model to obtain the user interest characteristics in the target platform, wherein the user interest analysis sample comprises the user interest characterization data and preset user interest characteristics corresponding to the user interest characterization data.

Fig. 4 is a construction structure diagram of a recommendation service and a search service provided in an embodiment of the present application, firstly, multimedia data is obtained, then content understanding is performed on the multimedia data to obtain multimedia content features, then a search service is constructed based on the multimedia content features and search words, the search service is deployed, and a recommendation service is constructed based on the multimedia content features and user interest features, and the recommendation service is deployed. After the recommendation service and the search service are deployed, the user generates new user interest characterization data by using the recommendation service and the search service, feeds the user interest characterization data back to the user interest model, generates new user interest characteristics by using the user interest model, reconstructs the recommendation service, and deploys the recommendation service. In addition, the multimedia data is continuously updated, and the content understanding process is continuously updated, so that the search service and the recommendation service are continuously updated and repeatedly performed, and finally the search service and the recommendation service reaching the preset effect are constructed.

Fig. 5 is a schematic diagram of a service construction process according to an embodiment of the present application, where the service construction process includes three stages: a cross-domain phase, a domain phase, and a customization phase, wherein:

The cross-domain stage mainly comprises the steps of obtaining multimedia data such as short videos, live broadcasting, atlases and the like in a plurality of domains, obtaining multimedia description information such as comment information, heat information and the like of each multimedia data, and adopting a process shown in fig. 4 to produce a general model such as a general content understanding model, a general user interest model, a general recommendation model and a general search model.

The domain stage mainly obtains multimedia data in a specific domain and multimedia description information of each multimedia data, and adopts a process production domain model shown in fig. 4 on the basis of a general model. Assuming that the generic model is generated using multimedia data for n fields, this stage can generate n field models, n being an integer greater than 1.

The customization stage mainly obtains the multimedia data and the multimedia description information of each multimedia data in a specific platform in a field, and generates a special model of the platform by adopting the process shown in fig. 4 on the basis of a field model in the field. How many platform customization services in the field can generate how many specialized models. That is, the values of i, j, k in fig. 5 are uncertain.

Fig. 6 is a schematic diagram of a customization process of a recommendation service provided in an embodiment of the present application. When a recommended service needs to be customized for the target platform, a technician of the target platform can provide the domain description information of the target platform, the service description information of the recommended service, data samples in the target platform required for customizing the recommended service, and optimization targets such as browsing duration and transaction amount of the recommended service to the server through the customized service interface.

In practical application, the data samples in the target platform required for customizing the recommendation service at least comprise: the user interest feature sample in the target platform and the preset multimedia data (predetermined by the technician according to experience) corresponding to the user interest feature sample. In addition, if it is desired to generate a dedicated recommendation model by means of a domain content understanding model and a domain user interest model, the data sample may further include: content understanding samples and user interest analysis samples in the target platform. The process of generating a dedicated recommendation model is subsequently described by way of example with the aid of a domain content understanding model and a domain user interest model. In addition, in order to ensure the accuracy of the sample, the labeling information of the user interest characteristic sample, the content understanding sample and the user interest analysis sample at this stage can be manually labeled or labeled by a pre-machine and then manually corrected.

Further, the server may determine a target domain to which the target platform belongs according to the domain description information, and then select a domain recommendation model providing a recommendation service corresponding to the service description information, a domain content understanding model providing a content understanding service, and a domain user interest model providing a user interest analysis service from the target domain.

And then, performing migration learning on the service relationship of the content understanding service in the target platform by utilizing the service relationship of the content understanding service in the target field and the content understanding sample in the target platform learned by the field content understanding model to obtain a special content understanding model of the target platform. And carrying out content understanding on each multimedia data in the target platform by utilizing the special content understanding model to obtain a multimedia feature library of the target platform.

And performing migration learning on the service relationship of the user interest analysis service in the target platform by using the service relationship of the user interest analysis service in the target domain learned by the domain user interest model and the user interest analysis sample in the target platform to obtain a special user interest model. And then, carrying out interest analysis on the user interest characterization data of each user in the target platform by using the special user interest model, so as to obtain a user interest feature library of the target platform.

And then, selecting some user interest feature samples from the user interest feature library of the target platform, and using the service relationship of recommended service in the target field learned by the field service model, such as the matching relationship between interest and content, the matching relationship between similar users and the matching relationship between similar content, and selecting the multimedia content features conforming to each user interest feature sample from the multimedia feature library of the target platform, wherein the multimedia data with the multimedia content features are used as candidate multimedia data.

Further, historical user operation data corresponding to each candidate multimedia data is analyzed, a first matching degree between the candidate multimedia data and an optimization target of the recommendation service is determined, and the matching degree between the candidate multimedia data and the optimization target of the recommendation service is determined according to the first matching degree. Sequencing the candidate multimedia data according to the sequence from high to low of the matching degree between the candidate multimedia data and the optimization target of the recommendation service, and adjusting model parameters of the field recommendation model according to the sequencing result and preset multimedia data corresponding to the current user interest characteristic sample until the recommendation accuracy of the field recommendation model is determined to reach a first set value, and determining the field recommendation model as a special recommendation model of the target platform. The recommendation service can be provided in the target platform by using the special recommendation model.

Subsequently, new user interest characterization data in the target platform can be obtained, and the special recommendation model is updated by utilizing the new user interest characterization data. Thus, the service accuracy of the special recommendation model is further improved.

In the embodiment of the application, the service relationships of the recommended services in different fields including the target field are learned in advance to obtain a universal recommendation model, and then the service relationships of the recommended services in different fields learned by the universal recommendation model are utilized to perform migration learning on the service relationships of the recommended services in the target field to obtain the field recommendation model. Subsequently, when customizing the recommended service for the target platform, selecting a domain recommended model from the target domain to which the target platform belongs, performing migration learning on the service relationship of the recommended service in the target platform by utilizing the service relationship of the recommended service in the target domain learned by the domain recommended model and the data sample in the target platform required by the customized recommended service to obtain a special recommended model of the target platform, and providing the recommended service in the target platform by utilizing the special recommended model.

Therefore, a developer can quickly customize the recommended service for the target platform without concerning the bottom implementation of the recommended service, and the development threshold of the recommended service can be greatly reduced. The general recommendation model learns the service relation of the recommended service in different fields, the field recommendation model continuously learns the service relation of the recommended service in the target field on the basis of the general recommendation model, the special recommendation model continuously learns the service relation of the recommended service in the target platform belonging to the target field on the basis of the field recommendation model, the process of gradually learning the service relation of the recommended service from easy to difficult and from thick to thin refines the recommended target of the recommendation model in each stage, the convergence speed of the recommendation model in each stage can be accelerated, and the machine learning rule is also more met, so that the convergence speed and the recommendation accuracy of the finally obtained special recommendation model can be improved.

In addition, in the embodiment of the application, the optimization target of the recommendation service in the target platform can be set, so that recommendation strategies of special recommendation models defined by different platforms in the target field are different, the recommendation models which better meet the service recommendation requirements of the target platform can be conveniently defined, and the user experience is better.

Fig. 7 is a schematic diagram of a customizing process of a search service provided in an embodiment of the present application. When it is desired to customize a search service for a target platform, a technician may provide domain description information of the target platform, service description information of the search service, data samples in the target platform required for customizing the search service, and optimization targets such as relevance and total amount of the search service to a server through a customized service interface.

Wherein, the data samples in the target platform required by the customized search service at least comprise: the method comprises the steps of searching a word feature sample of a target platform, preset multimedia data (predetermined by a technician according to experience) corresponding to the search word feature sample, and a user interest feature sample corresponding to the search word feature sample. In addition, if it is desired to understand the model by means of domain content, the data sample may further include: content in the target platform understands the sample. The process of generating a dedicated search model is described next by way of example with the aid of a domain content understanding model. In addition, in order to ensure the accuracy of the sample, the labeling information of the search word characteristic sample, the content understanding sample and the user interest analysis sample at the stage can be manually labeled or labeled by a pre-machine and then manually corrected.

Further, the server may determine a target domain to which the target platform belongs according to the domain description information, and then select a domain search model providing a search service corresponding to the service description information, a domain content understanding model providing a content understanding service, and a domain user interest model providing a user interest analysis service from the target domain.

And then, performing migration learning on the service relationship of the content understanding service in the target platform by utilizing the service relationship of the content understanding service in the target field and the content understanding sample in the target platform learned by the field content understanding model to obtain a special content understanding model of the target platform. And then, carrying out content understanding on each multimedia data in the target platform by utilizing the special content understanding model to obtain a multimedia feature library of the target platform.

And then, using the service relation of search service in the target field learned by the field service model, such as correlation between texts and correlation between semantics, selecting multimedia content characteristics conforming to each search word characteristic sample from a multimedia characteristic library of the target platform, and taking the multimedia data with the multimedia content characteristics as candidate multimedia data.

Further, historical user operation data corresponding to each candidate multimedia data are analyzed, a first matching degree between the candidate multimedia data and an optimization target of the search service is determined, the candidate service response data are compared with user interest feature samples corresponding to the search word feature samples, a second matching degree between the candidate service response data and the optimization target of the search service is determined, and then the matching degree between the candidate multimedia data and the optimization target of the search service is determined according to the first matching degree and the second matching degree. And then, sequencing the candidate multimedia data according to the sequence from high to low of the matching degree between the candidate multimedia data and the optimization target of the search service, and adjusting model parameters of the domain search model according to the sequencing result and preset multimedia data corresponding to the characteristic sample of the current search word until the search accuracy of the domain search model reaches a second set value, and determining the domain search model as a special search model of the target platform. The search service can be provided in the target platform by using the special search model.

Subsequently, new user interest characterization data in the target platform can be obtained, and the special search model is updated by utilizing the new user interest characterization data. Thus, the service accuracy of the special search model is further improved.

In the embodiment of the application, the service relations of the search services in different fields including the target field are learned in advance to obtain a general search model, and then the service relations of the search services in different fields learned by the general search model are utilized to perform migration learning on the service relations of the search services in the target field to obtain the field search model. Subsequently, when customizing the search service for the target platform, selecting a domain search model from the target domain to which the target platform belongs, performing migration learning on the service relationship of the search service in the target platform by utilizing the service relationship of the search service in the target domain learned by the domain search model and the data sample in the target platform required by the customized search service to obtain a special search model of the target platform, and providing the search service in the target platform by utilizing the special search model.

Therefore, a developer can rapidly customize the search service for the target platform without concerning the bottom implementation of the search service, and the development threshold of the search service can be greatly reduced. The general search model learns the service relation of the search service in different fields, the field search model continuously learns the service relation of the search service in the target field on the basis of the general search model, the special search model continuously learns the service relation of the search service in the target platform belonging to the target field on the basis of the field search model, the process of gradually learning the service relation of the search service from easy to difficult and from thick to thin refines the search targets of the search models in each stage, the convergence speed of the search models in each stage can be accelerated, and the machine learning rule is also more met, so that the convergence speed and the search accuracy of the finally obtained special search model can be improved.

In addition, in the embodiment of the application, the optimization target of the search service in the target platform can be set, so that the search strategies of the special search models defined by different platforms in the target field are different, the search models which better meet the service search requirements of the target platform can be conveniently defined, and the user experience is better.

In the embodiment of the application, 20 popular fields of multimedia data can be used as mixed training samples (graphics context/short video/live broadcast and the like) to train a general model such as a general content understanding model, a general user interest model, a general recommendation model and a general search model. Then, a domain knowledge base of 20 domains is built as a domain training sample of the domain, and domain models such as a domain content understanding model, a domain user interest model, a domain recommendation model and a domain search model are trained by using the domain training samples on the basis of a general model, wherein the domain knowledge base of each domain comprises an entity base of the domain, entity attributes, entity relations and consumption behavior portraits of users in the domain.

Subsequently, when the target platform is served, an application program interface (Application Program Interface, API) for docking the unique data of the target platform and the optimized parameter setting of the customized service are provided, after the multimedia data, the user interest characteristic characterization data and the optimized parameter of the customized service of the target platform are accessed, the domain model is used as a basis for transfer learning, a special model of the target platform such as a special recommendation model and a special search model is established, and the optimized target of the customized service is used for fine tuning the recommended service or the sequencing strategy of the search service, so that the deep customized service is provided for the target platform quickly.

The special model inherits the suitability of the universal model in the whole field, so that the calculation amount and the data iteration cost of the prior model training can be saved, the iteration work of the subsequent special model is focused on the optimization target of the customized service, namely the actual requirement of a customer, and the differentiated special model can be customized for different platforms correspondingly, thereby providing a set of solution capable of actively adapting to the search service and the recommendation service of the multimedia data in the current mainstream vertical industry.

The above process is described below in connection with specific embodiments.

The first stage: the whole network captures 1000 ten thousand pieces of multimedia data such as short videos and live streams, the 1000 ten thousand pieces of multimedia data are distributed in 20 popular fields, and a universal model such as a universal content understanding model, a universal user interest model, a universal recommendation model and a universal search model is trained by using the 1000 ten thousand pieces of multimedia data, wherein the universal recommendation model can recommend the 1000 ten thousand pieces of multimedia data indiscriminately, and the universal search model can search the 1000 ten thousand pieces of multimedia data indiscriminately.

Input and output data of each general model at this stage are exemplified:

1. a generic content understanding model.

Input: the multimedia description information of the multimedia data may be in the form of:

{ItemID：it10000001；

URL:……；

original content slice: a binary picture set;

original audio: voice separated from short video and live broadcast;

original text: text descriptions extracted from the web page;

}

and (3) outputting: the multimedia content features obtained through the content understanding model can be in the following form:

{ItemID：it10000001；

URL:……；

original content slice: a binary picture set;

and (3) voice recognition: words obtained by voice recognition from the short video;

character recognition: title, word assignment, caption, etc. identified from the image in the short video clip;

face identification: the face features identified from the short video/live broadcast/picture are represented by face identification;

and (3) tag: video classification of multimedia data such as news-domestic news, commodity-digital 3C, etc.;

entity identification: such as automobiles, scenic spots, commodities, animals, plants, etc.;

key frame: n frame-cut pictures which can most represent short video;

key frame visual features: low-dimensional computer visual features of key frames;

text keywords: content description text, word segmentation is carried out through natural language processing, and keywords after stop words and useless words are removed;

semantic Embedding: a low latitude spatial mapping representation of the short video;

…

}

2. General user interest model.

Input: behavior of the user group. User interest feature characterization data, such as comment pages, interest lists, favorites lists, forwarding relationships, etc., for each large website opening. Can be of the following form:

(Group feature)->(action feature)

(gender, city, age, cell phone brand, time) - > (item of merchandise in website a, certain merchandise, purchase comment content, merchandise category);

(gender, city, age, cell phone brand, time) - > (an article in B application, a tourist attraction, praise, travel);

(gender, city, age, cell phone brand, time) - > (one short video in C application, make-up self-timer content, praise, share, make-up)

And (3) outputting: the interest points and consumption contents of the user group can be in the following forms:

(gender, city, age, cell phone brand, time) - > (points of interest: probability), (consumer content identification).

3. A generic recommendation model.

Input: all content identifications after content understanding and all user group identifications after user interest analysis;

and (3) outputting: recommended short video/live/picture.

Generally, the ranking targets of the general recommendation model, such as Area Under Curve (Area Under Curve) and some technical indexes, are mainly used for measuring the relevance of the general recommendation model.

4. A generic search model.

Input: all content identifications after content understanding, all user group identifications after user interest analysis, and search term feature samples.

And (3) outputting: short video/live/picture searched.

In general, ranking targets such as text relevance and some technical indexes of a general search model are mainly used for measuring whether a search result is related to a keyword feature sample.

And a second stage: and directionally grabbing data of a plurality of automobile stations in the target field, such as the automobile field, and generating an automobile database, corresponding configuration drawings and comment information. And additionally selecting 10 tens of thousands of short videos and live broadcast contents in the automobile field for machine and/or manual annotation, so that the short videos and live broadcast can be associated with automobile models and automobile attributes. And aiming at the automobile field, the ranking strategy and the target of the recommended service and the search field can be finely adjusted, and the field model of the automobile field such as a field content understanding model, a field user interest model, a field recommended model and a field search model can be trained.

Input and output data of each general model at this stage are exemplified:

1. domain content understanding model.

Input: two inputs are added on the basis of a general content understanding model, wherein one is the expertise of the automobile field, and the other is a small number of data samples which are marked on the automobile field independently, such as adding detailed parameter information of each automobile.

For example, in 1000 ten thousand general multimedia data, if a car is in the multimedia data or a specific car type can be identified, the general 1000 ten thousand multimedia data can be directly connected with a car knowledge base, and at this time, the content understanding result of the multimedia data integrates all data information of the car and the corresponding relation between the multimedia data and the 10 ten thousand newly added multimedia data.

{ItemID：it10000001；

(middle part is not different from general content understanding model);

domain: an automobile;

brand: … …;

specific model: … …;

detailed characteristics: quotation, engine, length, width, height, power, interior decoration and color;

the associated content: (it 30000005, it80000004, it 90000002);

}

2. domain user interest model.

User interest characteristic characterization data matched with the field of 10 thousands of multimedia data, such as comments, praise, attention list and the like of website opening are directionally captured.

Input: the knowledge base of the additional field is added to help the general user interest model to understand more fine granularity, and the knowledge base can be in the following form: .

(gender, city, age, cell phone brand, time) - > (D application, comment, i prefer not too much XX, i feel that the same price buying XX is better);

(gender, city, age, cell phone brand, time) - > (E application, praise, XX test drive experience good);

and (3) outputting: the additional increase of the interest of the user group to the domain fine-grained entity, such as branding, model number, etc., can be in the following form: .

(gender, city, age, cell phone brand, time) - > (point of interest: probability), (consumption content identification) - > -car (brand: probability), (consumption car model);

3. the domain recommendation model.

The input of the field recommendation model is similar to that of the general recommendation model, but extra multimedia features in the automobile field are added compared with that of the general recommendation model, and the output of the field recommendation model is similar to that of the general recommendation model, but the popular brands in the automobile field can be recommended.

In general, the ranking goal of the generic recommendation model may employ domain goals of the automotive domain, such as browsing duration, brand attention, and so forth.

4. Domain search models.

The input of the domain search model is similar to that of the general search model, but extra multimedia features of the automobile domain are added compared with that of the general search model, and the output of the domain search model is similar to that of the general search model, but popular brands in the automobile domain can be searched.

In general, the ranking objective of the domain search model may employ experience rules of the automotive domain, such as automotive brand break-up, middle and outer financing priority, etc.

And a third stage: a certain platform applies for recommendation service and search service in the automobile field, and the recommendation service and the search service are quickly constructed by importing actual business content and user interest characteristic characterization data of the platform, so that automobile-related videos with granularity of automobile models can be recommended for users in the platform, and automobile-related videos with related automobile names can be searched.

In practical application, the actual service data comprises three parts:

a first part: stock quantity units (Stock Keeping Unit, SKU) library actually sold by the enterprise and corresponding detailed information.

Such as: commodity detailed information of 2000 automobile parts sold together, quotation, sales site, customer service and the like. For another example: some cars sell 10 commodities such as leather cushion, film, perfume, car cover and the like.

A second part: the method is used for recommending and searching picture and text evaluation, short video introduction, live broadcast playback and the like.

For example, 10 pictures and texts are written, 20 short videos are recorded and 10 live broadcast is started for the leather cushion of a certain car.

Third section: user behavior data at the account level.

Such as what the user has seen, the item detail page entered, the final order deal, order details, item attention, content praise, etc.

This stage may communicate with the technical team of the platform an optimization objective of explicit recommendation services and search services, e.g., optimization objective of recommendation services is browsing duration and total amount of deals, optimization objective of search model is relevance and total amount of deals.

After explicitly recommending the service or searching for the optimization objective of the service, the training samples of the ranking strategy in the service can be screened from the behavior data of the user by means of a machine algorithm.

For example, the optimization objective of the recommended service is browsing duration and total amount of deals, then:

the screening rules for positive samples may be: the total duration of browsing the image-text/short video/live broadcast content and the commodity detail page by the user exceeds 60s, and finally the user successfully orders.

The screening rules for negative samples may be: and the duration of browsing the image-text/short video/live broadcast content by the user is less than 30s, and finally, the user does not enter the commodity detail page and does not order.

Training the ranking strategy of the recommended service by using the new sample.

Here, if the optimization target of the recommendation service is browsing duration, 1 ten thousand multimedia data samples can be selected for manual correction and labeling, and the special recommendation model is trained by using the 1 ten thousand multimedia data samples. The special recommendation model can learn the relation between various automobiles and corresponding sales accessories, so that after a user browses short videos related to a certain automobile, the short videos of the corresponding automobile accessories in the platform can be automatically recommended to the user. In addition, as the browsing time length is set as the optimization target, the special user interest model can learn the consumption habit of the user, so that the parameters of the special user interest model are adjusted, the finally recommended multimedia data of the special recommendation model are more in line with the browsing behavior of the user, and the browsing time length is longer. These customizations are unique to the platform and are not available to other platforms in the automotive field for providing recommended services.

For another example, the optimization objective of the search service is relevance and total amount of deals, then:

the screening rules for positive samples may be: keyword hit measures that the search engine quality index (Discounted Cumulative Gain, DCG) score is high;

the screening rules for negative samples may be: the DCG score of the keyword hit is low, and finally, the item detail page is not ordered.

Then, the new sample is used to train the sorting strategy of the search service.

The optimization process for the search service is similar to the recommendation service and will not be described in detail herein.

When the method provided in the embodiments of the present application is implemented in software or hardware or a combination of software and hardware, a plurality of functional modules may be included in an electronic device, where each functional module may include software, hardware, or a combination thereof.

Fig. 8 is a schematic structural diagram of a service customization device according to an embodiment of the present application, which includes a receiving module 801, a determining module 802, a selecting module 803, a training module 804, and a service module 805.

A receiving module 801, configured to receive a service customization request, where the service customization request at least includes domain description information of a target platform, service description information of a customized service, and a data sample in the target platform required by the customized service;

A determining module 802, configured to determine, according to the domain description information, a target domain to which a target platform belongs;

a selection module 803, configured to select, from the target domain, a domain service model that provides a customized service corresponding to the service description information, where the domain service model is obtained by performing migration learning on a service relationship of the customized service in the target domain by using a service relationship of the customized service in a different domain including the target domain learned by a generic service model, where the service relationship of the customized service is a feature matching relationship between service input data and service response data of a specified uniform service;

the training module 804 is configured to perform migration learning on the service relationship of the customized service in the target platform by using the service relationship of the customized service in the target domain and the data sample learned by the domain service model, so as to obtain a dedicated service model of the target platform;

a service module 805, configured to provide a customized service in the target platform using the dedicated service model.

In one possible implementation, the training module 804 is specifically configured to:

Sorting the candidate service response data;

In one possible implementation, if the service customization request further includes an optimization objective of customizing a service, the training module 804 is specifically configured to:

the selecting module 803 is further configured to select a domain content understanding model that provides a content understanding service from the target domain, where the domain content understanding model is obtained by performing migration learning on a service relationship of the content understanding service in the target domain by using a service relationship of the content understanding service in a different domain including the target domain learned by a general content understanding model, where the service relationship of the content understanding service refers to a content understanding relationship between multimedia data and multimedia content features; and

the selecting module 803 is further configured to select a domain user interest model that provides a user interest analysis service from the target domain, where the domain user interest model is obtained by performing migration learning on a service relationship of the user interest analysis service in the target domain by using a service relationship of the user interest analysis service in a different domain including the target domain learned by a general user interest model, where the service relationship of the user interest analysis service refers to an association relationship between user interest characterization data and user interest features; and

In one possible embodiment, the method further comprises:

a feedback module 806, configured to obtain new user interest characterization data in the target platform after providing the customized service in the target platform using the dedicated service model; and updating the special service model by utilizing the new user interest characterization data.

The division of the modules in the embodiments of the present application is schematically only one logic function division, and there may be another division manner in actual implementation, and in addition, each functional module in each embodiment of the present application may be integrated in one processor, or may exist separately and physically, or two or more modules may be integrated in one module. The coupling of the individual modules to each other may be achieved by means of interfaces which are typically electrical communication interfaces, but it is not excluded that they may be mechanical interfaces or other forms of interfaces. Thus, the modules illustrated as separate components may or may not be physically separate, may be located in one place, or may be distributed in different locations on the same or different devices. The integrated modules may be implemented in hardware or in software functional modules.

Fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present application, where the electronic device includes a transceiver 901 and physical devices such as a processor 902, and the processor 902 may be a central processing unit (Central Processing Unit, CPU), a microprocessor, an application specific integrated circuit, a programmable logic circuit, a large-scale integrated circuit, or a digital processing unit. The transceiver 901 is used for transmitting and receiving data between the electronic device and other devices.

The electronic device may further comprise a memory 903 for storing software instructions to be executed by the processor 902, and of course some other data required by the electronic device, such as identification information of the electronic device, encryption information of the electronic device, user data, etc. The Memory 903 may be a Volatile Memory (RAM), such as Random-Access Memory (RAM); the Memory 903 may also be a Non-Volatile Memory (Non-Volatile Memory), such as Read-Only Memory (ROM), flash Memory (Flash Memory), hard Disk (HDD) or Solid State Drive (SSD), or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto. The memory 903 may be a combination of the above.

The specific connection medium between the processor 902, the memory 903, and the transceiver 901 is not limited in the embodiments of the present application. In the embodiment of the present application, the memory 903, the processor 902 and the transceiver 901 are only illustrated as being connected by a bus 904 in fig. 9, and the bus is indicated by a thick line in fig. 9, and the connection manner between other components is only illustrated schematically, but not limited to. The buses may be classified as address buses, data buses, control buses, etc. For ease of illustration, only one thick line is shown in fig. 9, but not only one bus or one type of bus.

The processor 902 may be dedicated hardware or a processor running software, and when the processor 902 can run software, the processor 902 reads the software instructions stored in the memory 903 and performs the service customization method referred to in the foregoing embodiment under the drive of the software instructions.

The present application also provides a storage medium, which when executed by a processor of an electronic device, is capable of executing the service customization method referred to in the foregoing embodiment.

In some possible embodiments, aspects of the service customization method provided in the present application may also be implemented in the form of a program product, where the program product includes program code for causing an electronic device to perform the service customization method as referred to in the foregoing embodiments, when the program product is run on the electronic device.

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

The program product for service customization in embodiments of the present application may take the form of a CD-ROM and include program code that can run on a computing device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

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

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

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

It should be noted that although several units or sub-units of the apparatus are mentioned in the above detailed description, such a division is merely exemplary and not mandatory. Indeed, the features and functions of two or more of the elements described above may be embodied in one element in accordance with embodiments of the present application. Conversely, the features and functions of one unit described above may be further divided into a plurality of units to be embodied.

Furthermore, although the operations of the methods of the present application are depicted in the drawings in a particular order, this is not required to or suggested that these operations must be performed in this particular order or that all of the illustrated operations must be performed in order to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform.

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

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. 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.

While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the application.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present application without departing from the spirit or scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims and the equivalents thereof, the present application is intended to cover such modifications and variations.

Claims

1. A method of customizing a service, comprising:

providing a customized service in the target platform using the dedicated service model;

Performing migration learning on the service relationship of the customized service in the target platform by using the service relationship of the customized service in the target domain and the data sample learned by the domain service model to obtain a special service model of the target platform, wherein the method comprises the following steps:

sorting the candidate service response data;

2. The method of claim 1, wherein sorting the candidate service response data if the service customization request further includes an optimization objective for customizing the service, comprises:

3. The method of claim 2, wherein calculating a degree of matching between each candidate service response data and an optimization objective of the customized service comprises:

4. The method of claim 1, wherein when the customized service is a recommended service or a search service, the service response data of the customized service is a multimedia content feature, and further comprising:

5. The method of claim 1 or 4, wherein when the customized service is a recommended service, service input data of the customized service is a user interest feature; when the customized service is a search service, the service input data of the customized service is a search word feature sample, the search word feature sample corresponding to a user interest feature, and further comprising:

6. The method of any of claims 1-4, further comprising, after providing a customized service in the target platform using the dedicated service model:

acquiring new user interest characterization data in the target platform;

7. A service customization device, comprising:

the service module is used for providing customized service in the target platform by utilizing the special service model;

the training module is specifically configured to screen candidate service response data from a service response database of the target platform by using the service relationship of the customized service in the target domain and the service input data sample of the customized service in the target platform, which are learned by the domain service model; sorting the candidate service response data; and adjusting model parameters of the domain service model according to the sorting result and preset service response data corresponding to the service input data sample until the service accuracy of the domain service model is determined to reach a set value, and determining the domain service model as a special service model of the target platform.

8. The apparatus of claim 7, wherein if the service customization request further includes an optimization objective for customizing a service, the training module is specifically configured to:

9. The apparatus of claim 8, wherein the training module is specifically configured to:

10. The apparatus of claim 7, wherein when the customized service is a recommended service or a search service, the service response data of the customized service is a multimedia content feature;

11. The apparatus of claim 7 or 10, wherein when the customized service is a recommended service, the service input data of the customized service is a user interest feature; when the customized service is a search service, service input data of the customized service is a search word feature sample, and the search word feature sample corresponds to a user interest feature;

12. The apparatus of any one of claims 7-10, further comprising:

13. An electronic device, comprising: at least one processor, and a memory communicatively coupled to the at least one processor, wherein:

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

14. A storage medium, characterized in that the electronic device is capable of performing the method of any of claims 1-6 when instructions in the storage medium are executed by a processor of the electronic device.