CN112214682B - Recommendation method, device and equipment based on field and storage medium - Google Patents

Abstract

The embodiment of the application discloses a recommendation method, a device, equipment and a storage medium based on the field, wherein the method comprises the following steps: acquiring target information of a target domain, source domain information of all source domains, characteristic information of target items in the target domain and characteristic information of source items in all source domains on a terminal; determining the priority sequence of the recommendation modes based on the field according to the target information, the source domain information, the feature information of the target item and the feature information of the source item; according to the priority ranking, determining recommended items by sequentially selecting corresponding recommendation modes to form a recommendation list corresponding to the terminal; and sequentially sending the recommended items to the terminal according to the recommendation list. Therefore, the recommendation accuracy can be improved, and a more complete and comprehensive recommendation list can be obtained.

Description

Recommendation method, device and equipment based on field and storage medium

Technical Field

The embodiment of the application relates to the technical field of big data, and relates to but is not limited to a recommendation method, a recommendation device, recommendation equipment and a storage medium based on the field.

Background

The cross-domain recommendation technology combines data in a plurality of domains and takes the data into consideration, and performs recommendation by jointly acting on a target domain through a certain relation. For example, if a domain has user-book rating data and another domain has user-movie rating data, then when a cross-domain recommendation technique is applied to recommend books to a user, not only the user-book rating data is considered, but also the user-movie rating data of the user in the movie domain and other information which can be obtained and is possibly beneficial to the recommendation of the user in the book domain are considered, and then personalized recommendations are provided for the user in the domain.

The current cross-domain recommendation technology generally crawls a large amount of user interaction data on numerous APPs of a terminal through a web crawler, and then preprocesses the interaction data; the data is then analyzed by a recommendation engine to generate a recommendation list.

However, the current cross-domain recommendation technology is to complete a recommendation task in a specific recommendation manner for a certain scene, which obviously reduces the recommendation accuracy.

Disclosure of Invention

In view of this, the present application provides a recommendation method, apparatus, device and storage medium based on a domain.

The technical scheme of the embodiment of the application is realized as follows:

in a first aspect, an embodiment of the present application provides a recommendation method based on a domain, where the method includes:

acquiring target information of a target domain, source domain information of all source domains, characteristic information of target items in the target domain and characteristic information of source items in all source domains on a terminal;

determining the priority ranking of the recommendation modes based on the field according to the target information, the source domain information, the feature information of the target item and the feature information of the source item;

according to the priority ranking, determining recommended items by sequentially selecting corresponding recommendation modes to form a recommendation list corresponding to the terminal;

and sequentially sending the recommended items to the terminal according to the recommended list.

In a second aspect, an embodiment of the present application provides a recommendation device based on a domain, where the device includes:

an acquiring unit, configured to acquire target information of a target domain on a terminal, source domain information of all source domains, feature information of a target item in the target domain, and feature information of source items in all source domains;

the determining unit is used for determining the priority sequence of the recommendation modes based on the field according to the target information, the source domain information, the feature information of the target item and the feature information of the source item;

the processing unit is used for sequentially selecting corresponding recommendation modes to determine recommendation items according to the priority ranking so as to form a recommendation list corresponding to the terminal;

and the sending unit is used for sequentially sending the recommended items to the terminal according to the recommended list.

In a third aspect, an embodiment of the present application provides a recommendation device based on a domain, where the device at least includes: a processor and a storage medium configured to store executable instructions, wherein: the processor is configured to execute stored executable instructions; the executable instructions are configured to perform the domain-based recommendation method described above.

In a third aspect, an embodiment of the present application provides a storage medium, where computer-executable instructions are stored in the storage medium, and the computer-executable instructions are configured to execute the foregoing domain-based recommendation method.

According to the recommendation method, device, equipment and storage medium based on the field, the priority ranking of the recommendation mode based on the field is determined according to the target information, the source domain information, the feature information of the target item and the feature information of the source item; and determining recommended items by sequentially selecting corresponding recommendation modes according to the priority sequence so as to form a recommendation list corresponding to the terminal. Therefore, the plurality of recommendation modes based on the fields are subjected to priority ranking, the corresponding recommendation modes are selected to determine the recommendation items according to the priority ranking in sequence, the recommendation mode which can determine the recommendation items most accurately can be determined to be higher in priority to perform recommendation preferentially, other recommendation modes are considered, the recommendation accuracy can be improved, and a more complete and comprehensive recommendation list can be obtained.

Drawings

In the drawings, which are not necessarily drawn to scale, like reference numerals may describe similar components in different views. Like reference numerals having different letter suffixes may represent different examples of similar components. The drawings illustrate generally, by way of example, but not by way of limitation, various embodiments discussed herein.

Fig. 1 is a schematic diagram of an implementation process of cross-domain knowledge point combination according to an embodiment of the present application;

fig. 2 is a schematic diagram of another implementation process of combining cross-domain knowledge points according to an embodiment of the present application;

FIG. 3 is a flow chart illustrating an implementation of a domain-based recommendation method provided in the related art;

FIG. 4 is a schematic diagram illustrating an implementation flow of a domain-based recommendation method according to an embodiment of the present application;

FIG. 5 is a schematic diagram illustrating an implementation flow of a domain-based recommendation method according to an embodiment of the present application;

FIG. 6 is a schematic diagram illustrating an implementation flow of a domain-based recommendation method according to an embodiment of the present application;

FIG. 7 is a flowchart illustrating an implementation of a domain-based recommendation method according to an embodiment of the present application;

FIG. 8A is a diagram illustrating an implementation process of a multi-domain recommendation method;

FIG. 8B is a diagram illustrating an implementation process of a multi-connection recommendation method;

FIG. 8C is a schematic diagram illustrating an implementation process of a cross-domain recommendation method;

FIG. 9 is a flowchart illustrating an implementation of a domain-based recommendation method according to an embodiment of the present application;

FIG. 10 is a flowchart illustrating an implementation of a task priority based recommendation method according to an embodiment of the present application;

fig. 11 is a schematic flow chart illustrating an implementation of a domain-based recommendation method according to an embodiment of the present application;

FIG. 12A is a simplified flow diagram of a multi-domain recommendation;

FIG. 12B is a simplified flow diagram of a multi-connection recommendation;

FIG. 12C is a simplified flow diagram of a cross-domain recommendation;

fig. 13 is a schematic structural diagram illustrating a composition of a domain-based recommendation device according to an embodiment of the present application;

fig. 14 is a schematic structural diagram of a component of a recommendation device based on a domain according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, specific technical solutions of the present application will be described in further detail below with reference to the accompanying drawings in the embodiments of the present application. The following examples are intended to illustrate the present application but are not intended to limit the scope of the present application.

In the following description, suffixes such as "module", "component", or "unit" used to denote elements are used only for the convenience of description of the present application, and have no specific meaning by themselves. Thus, "module", "component" or "unit" may be used mixedly.

The cross-domain recommendation technology is to combine data in multiple domains and take the data into consideration, and perform recommendation by jointly acting on a target domain (i.e., target domain) through a certain relation. In general, the domain may be defined from several levels:

attribute level: the fields to which the items belong can be divided according to the attributes of the items, and the items with the same attribute are divided into the same field. For example, the attributes of a movie may include action and comedy, so that action is one domain and comedy is another domain, i.e., action and comedy are two different domains.

Type class: the fields to which the items belong can be divided according to the types of the items, and the items of the same type are divided into the same field. For example, movies and television shows are of different genres, and thus, movies and television shows are of different areas at the genre level.

Item level: the domain to which the item belongs can be divided according to the characteristics of the recommended item, and generally, the recommended item is different in type and most or all of the attributes are different. For example, movies and books belong to two different domains, even if they have some common attributes (e.g., title/publication date are the same).

System level: the domain of the item may be divided according to the system described by the item, i.e. the recommendations of items belonging to different systems considered as different domains. For example, movies viewed on Netflix and movies scored in the Movie Lens recommendation system are movies belonging to different systems and thus belong to different domains.

Here, a description is given to a method for combining cross-domain knowledge points, wherein the method for combining cross-domain knowledge points is mainly classified into the following two types:

and (3) knowledge aggregation: it is referred to collecting knowledge to form a knowledge cluster, so that through the knowledge cluster, it is possible to help a target domain generate a recommended item by the knowledge collected from a plurality of source domains.

As shown in fig. 1, a schematic diagram of an implementation process of combining cross-domain knowledge points provided in the embodiment of the present application is obtained by collecting knowledge in a source domain 101 and a target domain 102, performing aggregation processing 103 on the collected knowledge to form a knowledge cluster, and generating a recommendation item 104 for the target domain through the knowledge cluster.

Knowledge connection and knowledge migration: knowledge linking, meaning the mutual use of knowledge in different domains; the knowledge migration is to migrate the knowledge in the source domain to the target domain, and then recommend the target domain by the migrated knowledge. Namely knowledge linking and knowledge migration, enables knowledge to be migrated and linked between different domains to support recommendation of a target domain.

As shown in fig. 2, another implementation process diagram for combining cross-domain knowledge points is provided in the embodiment of the present application, and the knowledge in the source domain 201 is migrated to the target domain 202 through a migration process 203, so as to generate a recommendation item 204 for the target domain.

Fig. 3 is a schematic flow chart illustrating an implementation process of a domain-based recommendation method provided in the related art, and as shown in fig. 3, the domain-based recommendation method in the related art includes the following steps:

step S301, crawling a large amount of user interaction information on numerous Application programs (APP) of the terminal through a web crawler, and collecting data.

Step S302, data preprocessing is carried out on the collected data.

And step S303, storing the preprocessed data into a storage system.

Step S304, a preliminary recommendation list is generated by analyzing the data through a recommendation engine.

The recommendation engine analyzes data, which is mainly to recommend items to a target domain respectively according to any recommendation mode through data of various source domains and target domains, and mainly adopts methods of knowledge aggregation, knowledge connection and migration to complete a recommendation task.

And S305, performing candidate and sequencing on the preliminary recommendation list to perform recommendation result processing to form a final recommendation list, and sending a recommendation item to a user interaction interface of the terminal according to the recommendation list.

However, the domain-based recommendation method provided by the related art has at least the following problems:

firstly, in the related technology, the interactive information of most users on the APP on the terminal is crawled, and then a recommendation list is formed based on any recommendation mode, so that the recommendation experience functions of some users and the APPs of the kids can be ignored. For example, some users do not use the popular APP, but use a small APP, so that various interaction information of the users in the software cannot be collected, thereby influencing the description of the character image of the users.

Second, the recommendation method in the related art does not perform priority ranking on multiple recommendation modes, but only completes a certain recommendation mode for a certain scenario, that is, completes a recommendation task by using a specific recommendation mode, which obviously reduces the recommendation accuracy.

Thirdly, the particularity of the terminal recommendation system and the scene information of the terminal are not fully utilized, and the scene information can supplement the portrait presence information of the user.

Fourth, the function of updating the recommendation list in real time is not implemented.

Based on at least one of the problems in the related art, the embodiment of the application provides a field-based recommendation method, user portrait is supplemented by obtaining contextual data provided by a terminal, then all APP interaction information of the terminal is collected, mainly aiming at including interaction information of the APP of the Xiaozhong people used by the terminal to obtain personalized user portrait, then a recommendation list is formed by adopting deep learning and a recommendation method based on task priority, and is updated in real time, so that the recommendation accuracy can be improved, and a more complete and comprehensive recommendation list can be obtained.

Fig. 4 is a schematic implementation flow diagram of a domain-based recommendation method provided in an embodiment of the present application, and as shown in fig. 4, the method includes the following steps:

step S401, acquiring target information of a target domain, source domain information of all source domains, feature information of a target item in the target domain and feature information of source items in all source domains on a terminal.

The recommendation method based on the field in the embodiment of the present application may correspond to one terminal and may also correspond to multiple terminals, where the embodiment of the present application is described by taking one terminal as an example, the one terminal includes multiple APPs, where the multiple APPs may include popular APPs and also include popular APPs personally used by a user of the terminal.

The target information of the target domain is profile information of the target domain, for example, the target domain may be a target APP on the terminal, and then, the target information is software profile information of the target APP.

The target domain comprises one or more target items, the characteristic information of each target item is the profile information of the corresponding target item, when the target domain is a target APP, the target item can be a recommended item or a function item in the target APP, and the characteristic information of the target item can be the profile information of the recommended item or the function item. For example, when the target domain is a video playing APP, the target item may be a recommended movie, and the feature information of the target item may be profile information such as an introduction and a movie review of the movie.

The source domain information of the source domain is profile information of the source domain, for example, the source domain may be other APPs on the terminal, and then, the source domain information is software profile information of the other APPs.

The source domain also includes one or more source items, the feature information of each source item is profile information of the corresponding source item, when the source domain is another APP, the source item may be a recommended item or a function item in the other APP, and the feature information of the source item may be profile information of the recommended item or the function item. For example, when the source domain is a shopping APP, the source item may be recommended for an item, and the characteristic information of the source item may be profile information such as introduction of the item and comments of other buyers.

In the embodiment of the application, target information of a target domain, source domain information of all source domains, feature information of a target item in the target domain and feature information of source items in all source domains on the terminal can be crawled through a web crawler.

And S402, determining the priority sequence of the recommendation modes based on the field according to the target information, the source domain information, the feature information of the target item and the feature information of the source item.

Here, the domain-based recommendation method is one or more, and when there are a plurality of recommendation methods, it is necessary to determine a priority of each recommendation method in the plurality of recommendation methods, form a priority ranking of the plurality of recommendation methods, and implement recommendation by sequentially executing corresponding recommendation methods according to the priority ranking.

In the embodiment of the application, according to the target information and the source domain information, whether a source domain similar to the target domain exists on the terminal can be determined; according to the feature information of the target item and the feature information of the source item, whether more overlapped information exists between the target domain and the source domain can be determined.

If the source domain similar to the target domain exists on the terminal, the multi-domain recommendation mode can be preferentially executed, namely, the multi-domain recommendation mode is determined to have the highest priority; after the multi-domain recommendation mode is executed, continuing to judge the target domain, determining whether more overlapped information exists between the target domain and the source domain, if so, preferentially executing the multi-connection recommendation mode, and then executing the cross-domain recommendation mode, namely determining that the priority of the multi-connection recommendation mode is higher than that of the cross-domain recommendation mode; otherwise, if the existing cross-domain recommendation modes do not exist, the cross-domain recommendation modes are preferentially executed, and then the multi-connection recommendation modes are executed, namely the priority of the cross-domain recommendation modes is higher than that of the multi-connection recommendation modes.

If the source domain similar to the target domain does not exist on the terminal, continuing to judge the target domain, determining whether more overlapped information exists between the target domain and the source domain, if so, preferentially executing a multi-connection recommendation mode, and then executing a cross-domain recommendation mode, namely determining that the priority of the multi-connection recommendation mode is higher than that of the cross-domain recommendation mode; otherwise, if the existing cross-domain recommendation modes do not exist, the cross-domain recommendation modes are preferentially executed, and then the multi-connection recommendation modes are executed, namely the priority of the cross-domain recommendation modes is higher than that of the multi-connection recommendation modes.

And S403, determining recommended items by sequentially selecting corresponding recommendation modes according to the priority sequence to form a recommendation list corresponding to the terminal.

After the priority of each recommendation mode is determined, the corresponding recommendation modes are selected in sequence from high priority to low priority for recommendation according to the priority ranking of the recommendation modes.

For example, if it is determined that the priority of the multi-domain recommendation mode is the highest, the priority of the multi-connection recommendation mode is the second, and the priority of the cross-domain recommendation mode is the lowest, then the multi-domain recommendation mode, the multi-connection recommendation mode, and the cross-domain recommendation mode are sequentially recommended, and a recommendation list formed after the three recommendation modes are completed is determined as a recommendation list corresponding to the terminal.

The recommendation list comprises at least one recommendation item, the arrangement sequence of the recommendation items in the recommendation list is the sequence of the recommendation items formed when multiple recommendation modes are recommended, namely when multiple domain recommendation modes, multiple connection recommendation modes and cross domain recommendation modes are recommended in sequence, the arrangement sequence of the recommendation items in the recommendation list is the items recommended by the multiple domain recommendation modes, the items recommended by the multiple connection recommendation modes and the items recommended by the cross domain recommendation modes.

And S404, sequentially sending the recommended items to the terminal according to the recommended list.

After the recommendation list is determined, the recommendation items are sequentially sent to the terminal according to the arrangement sequence of the recommendation items in the recommendation list, and item recommendation is achieved for the terminal. In the implementation process, one recommended item may be sent at a time, that is, one recommended item is presented on the interactive page of the terminal at a time, and of course, multiple recommended items may also be sent at a time, that is, multiple recommended items are presented on the interactive page of the terminal at the same time, and at this time, the arrangement order of the multiple recommended items is the same as the arrangement order of the multiple recommended items in the recommendation list.

According to the field-based recommendation method provided by the embodiment of the application, the priority ranking of the field-based recommendation mode is determined according to the target information, the source domain information, the feature information of the target item and the feature information of the source item; and determining recommended items by sequentially selecting corresponding recommendation modes according to the priority ranking to form a recommendation list of the corresponding terminal. Therefore, the plurality of recommendation modes based on the fields are subjected to priority ranking, the corresponding recommendation modes are selected to determine the recommendation items according to the priority ranking in sequence, the recommendation mode which can determine the recommendation items most accurately can be determined to be the higher priority, so that the recommendation is performed preferentially, and the recommendation accuracy can be improved. Meanwhile, other recommendation modes are also considered, so that a more complete and comprehensive recommendation list can be obtained.

Fig. 5 is a schematic flow chart illustrating an implementation process of a domain-based recommendation method provided in an embodiment of the present application, where as shown in fig. 5, the method includes the following steps:

step S501, acquiring target information of a target domain, source domain information of all source domains, feature information of a target item in the target domain and feature information of source items in all source domains on a terminal.

Step S502, when the first similarity between the target information and the source domain information of any source domain is greater than or equal to a first preset threshold, determining that the multi-domain recommendation mode has a first priority.

Here, when a first similarity between the target information and source domain information of any one source domain is greater than or equal to a first preset threshold, it is determined that a source domain similar to the target domain exists on the terminal.

In the embodiment of the present application, the method for determining whether a source domain similar to the target domain exists on the terminal may be implemented by the following steps:

step S5021, comparing the target information with the source domain information of each source domain, and determining a first similarity between the target information and the source domain information of each source domain.

Here, the determining of the first similarity may be performed in any one of the following manners: a semantic understanding-based calculation method, a vector space model-based calculation method and a hamming distance-based calculation method.

Step S5022, determining whether at least one first similarity value is greater than or equal to a first preset threshold, and if so, determining that a source domain similar to the target domain exists on the terminal.

Here, the first preset threshold may be set according to the actual situation, and the first preset threshold may be a percentage value, and is used for determining the first similarity. For example, the first preset threshold may be 95%, and then, when at least one first similarity value is greater than or equal to 95%, it is determined that a source domain similar to the target domain exists on the terminal.

Step S5023, when a source domain similar to the target domain exists on the terminal, determining a multi-domain recommendation mode as having a first priority.

Here, the first priority is the highest priority, and when the multi-domain recommendation is determined to have the first priority, the multi-domain recommendation will be selected first for recommendation.

Step S503, when the number of the first source items in any source domain meets the preset condition, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority.

Here, the second similarity between the feature information of the first source item and the feature information of the target item is greater than or equal to a second preset threshold. When the number of the first source items in any source domain meets a preset condition, the target domain and the source domain have more overlapping information.

In the embodiment of the present application, the method for determining whether there is more overlapping information between the target domain and the source domain may be implemented by the following steps:

step S5031, determining a second similarity between the feature information of the source item in each source domain and the feature information of the target item.

Here, the determining of the second similarity may be performed by any one of: cosine similarity, pearson correlation coefficient, euclidean distance, and Jacard coefficient.

In step S5032, if the second similarity corresponding to any source item is greater than or equal to a second preset threshold, the corresponding source item is determined as the first source item.

Here, the second preset threshold may be set according to the actual situation, and the second preset threshold may also be a percentage value, which is used for determining the second similarity. For example, the second preset threshold may be set to 95%, and then, when the second similarity corresponding to any source item is greater than or equal to 95%, the corresponding source item is determined as the first source item.

Step S5033, when there is a first source item in any source domain and the number of the first source items meets a preset condition, determining that there is more overlapping information between the target domain and the source domain.

Step S5034, when it is determined that there is more overlapping information between the target domain and the source domain, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority.

Here, the first priority is higher than the second priority; the second priority is higher than the third priority. That is, if it is determined that there is more overlapping information between the target domain and the source domain, it is determined that the priorities of the multiple recommendation modes are ranked as a multi-domain recommendation mode, a multi-connection recommendation mode, and a cross-domain recommendation mode.

It should be noted that the first similarity generally refers to a similarity between target information and source domain information of any source domain, and does not particularly refer to a similarity between certain determined target information and certain source domain information; the second similarity generally refers to a similarity between the feature information of the first source item and the feature information of the target item, and does not particularly refer to a similarity between the feature information of a certain first source item and the feature information of a certain target item.

Step S504, when the number of the first source items in all the source domains does not meet the preset condition, determining that the cross-domain recommendation mode has a second priority, and the multi-connection recommendation mode has a third priority.

Here, when the number of the first source items in all the source domains does not satisfy the preset condition, it indicates that there is no more overlapping information between the target domain and the source domain, and then preferentially executing a cross-domain recommendation manner, that is, determining that the cross-domain recommendation manner has a second priority, the multi-connection recommendation manner has a third priority, and the second priority is higher than the third priority.

And step S505, according to the priority sequence, sequentially selecting corresponding recommendation modes to determine recommendation items so as to form a recommendation list corresponding to the terminal.

And step S506, sequentially sending the recommended items to the terminal according to the recommended list.

According to the field-based recommendation method provided by the embodiment of the application, whether a source domain similar to the target domain exists on the terminal is determined according to the target information and the source domain information; and determining whether more overlapped information exists between the target domain and the source domain according to the characteristic information of the target item and the characteristic information of the source item, thereby determining the priority ranking of the recommendation modes based on the domains. Therefore, the plurality of recommendation modes based on the field are subjected to priority ranking, the corresponding recommendation modes are selected to determine the recommendation items according to the priority ranking in sequence, the recommendation mode which can determine the recommendation items most accurately can be determined to be higher in priority, and therefore recommendation is performed preferentially, and the accuracy of recommendation can be improved. Meanwhile, other recommendation modes are also considered, so that a more complete and comprehensive recommendation list can be obtained. In addition, the source domain information can contain the APP of the crowd on the terminal due to the consideration of all the source domains on the terminal, so that a more accurate personalized user representation of the terminal user can be obtained, and more accurate recommended items can be recommended for the terminal.

Fig. 6 is a schematic implementation flow diagram of a domain-based recommendation method provided in an embodiment of the present application, and as shown in fig. 6, the method includes the following steps:

step S601, acquiring target information of a target domain on a terminal, source domain information of all source domains, feature information of a target item in the target domain, and feature information of source items in all source domains.

Step S602, when the first similarity between the target information and the source domain information of each source domain is smaller than a first preset threshold and the number of the first source items in any source domain meets a preset condition, it is determined that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority.

Here, when the first similarity between the target information and the source domain information of each source domain is less than a first preset threshold, it is determined that there is no source domain similar to the target domain on the terminal. In the embodiment of the application, if it is determined that the source domain similar to the target domain does not exist on the terminal, the target domain and the source domain on the terminal need to be continuously judged, and whether more overlapped information exists between the target domain and the source domain is determined.

Here, whether there is more overlapping information of the target domain and the source domain may be determined by:

step S6021, determining a second similarity between the feature information of the source item and the feature information of the target item in each source domain.

Step S6022, if the second similarity corresponding to any source item is greater than or equal to a second preset threshold, determining the corresponding source item as the first source item.

And the second similarity between the characteristic information of the first source item and the characteristic information of the target item is greater than or equal to a second preset threshold value.

Step S6023, when there is a first source item in any source domain and the number of the first source items meets a preset condition, determining that there is more overlapping information between the target domain and the source domain.

Step S6024, when it is determined that the target domain and the source domain have more overlapping information, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority.

Here, the second priority is higher than the third priority.

Step S603, when the number of the first source items in all the source domains does not satisfy the preset condition, determining that the cross-domain recommendation mode has a second priority, and the multi-connection recommendation mode has a third priority.

Here, when the number of the first source items in all the source domains does not satisfy the preset condition, it indicates that there is no more overlapping information between the target domain and the source domain, and then a cross-domain recommendation manner is preferentially executed.

And step S604, determining recommended items by sequentially selecting corresponding recommendation modes according to the priority sequence so as to form a recommendation list corresponding to the terminal.

And step S605, sequentially sending the recommended items to the terminal according to the recommended list.

According to the field-based recommendation method provided by the embodiment of the application, the priority ranking is performed on the plurality of field-based recommendation modes, the corresponding recommendation modes are selected to determine the recommended items according to the priority ranking in sequence, the recommendation mode which can determine the recommended items most accurately can be determined to be higher priority, so that the recommended items can be preferentially recommended, and the recommendation accuracy can be further improved.

Fig. 7 is a schematic implementation flow diagram of a domain-based recommendation method provided in an embodiment of the present application, and as shown in fig. 7, the method includes the following steps:

step S701, acquiring target information of a target domain, source domain information of all source domains, feature information of a target item in the target domain and feature information of source items in all source domains on a terminal.

Step S702, determining the priority sequence of the recommendation modes based on the field according to the target information, the source domain information, the feature information of the target item and the feature information of the source item.

And step S703, determining recommended items by sequentially selecting corresponding recommendation modes according to the priority ranking.

Step S704, when the recommendation mode selected according to the priority ranking is a multi-field recommendation mode, determining the items in the source domain having the same user with the target domain as the recommendation items to form a recommendation list corresponding to the terminal.

Here, the multi-domain recommendation method refers to recommending items of a source domain and a target domain to a user in the source domain and a user in the target domain. The recommendation method is mainly applied to cross-system level recommendation and requires user overlap between a source domain and a target domain.

As shown in fig. 8A, which is a schematic diagram of an implementation process of a multi-domain recommendation method, a user in a source domain is U _S The user in the target domain is U _T The item of the source domain is I _S Of the target domainItem is I _T . In the multi-domain recommendation method, there is user overlap between the source domain and the target domain, and items in the source domain and the target domain can be recommended to each other according to the source domain data 801 in the source domain and the target domain data 802 in the target domain, where the region shown by the dotted line is the user U in the source domain _S Or user U in the target domain _T A recommended item.

Step S705, when the recommendation mode selected according to the priority ranking is a multi-connection recommendation mode, adding the source domain information of the source domain to the target information of the target domain to determine recommendation items according to the added source domain information, and forming a recommendation list corresponding to the terminal.

Here, the multi-connection recommendation method is to enrich available knowledge of the target domain using source domain information collected by the source domain, thereby improving a recommendation rate of items in the target domain. The recommendation method requires that a relationship is generated between data and an overlap between a source domain and a target domain, and further an explicit or implicit knowledge-based connection between the source domain and the target domain is established.

As shown in fig. 8B, the implementation process of the multi-connection recommendation method is schematically illustrated, in the multi-connection recommendation method, the area shown by the dotted line is the user U in the target domain _T A recommended item.

Step S706, when the recommendation mode selected according to the priority ranking is a cross-domain recommendation mode, determining the items in the source domain belonging to different domains from the target domain as the recommendation items to form a recommendation list corresponding to the terminal.

Here, the cross-domain recommendation method may recommend items of the target domain to the user of the source domain through knowledge of the target domain. The recommendation method aims to provide recommendations for users without information in the target domain, in this case, data relationships and overlaps between domains do not exist between the domains, and it is necessary to establish a connection of relationships between the source domain and the target domain and migrate knowledge from the source domain to the target domain.

As shown in fig. 8C, a schematic diagram of an implementation process of a cross-domain recommendation manner is shown, in the cross-domain recommendation manner, an area shown by a dotted line is a user U in a source domain _S A recommended item.

And step S707, sequentially sending the recommended items to the terminal according to the recommendation list.

According to the field-based recommendation method provided by the embodiment of the application, the multi-field recommendation mode, the multi-connection recommendation mode and the cross-field recommendation mode are respectively selected to determine the recommendation items so as to form the recommendation list corresponding to the terminal, so that various recommendation modes can be considered simultaneously, and a more complete and comprehensive recommendation list can be obtained.

Fig. 9 is a schematic flow chart illustrating an implementation process of a domain-based recommendation method provided in an embodiment of the present application, and as shown in fig. 9, the method includes the following steps:

step S901, obtaining the scene data of the terminal.

Here, the scene data includes data such as location information, call frequency, power-on duration, power-on time, and terminal use frequency of the terminal.

Step S902, updating feature information of the target item in the target domain according to the scenario data.

Here, the contextual data may be added to the feature information of the target item, so that recommendation may be performed based on the contextual data at the same time in a subsequent recommendation process. For example, when the user carries the terminal to travel abroad, the position information of the terminal changes, and then when the group purchase APP is recommended, local food and tourist attractions can be recommended for the user according to the current position information.

In the embodiment of the present application, after the feature information of the target item is updated, the updated feature information of the target item is formed.

Step S903, acquiring the target information of the target domain on the terminal, the source domain information of all the source domains, the updated feature information of the target item in the target domain and the feature information of the source items in all the source domains.

Step S904, pre-processing the target information of the target domain and the source domain information of all the source domains to filter out invalid information in the target information and the source domain information.

The preprocessing of the target information of the target domain and the source domain information of all the source domains is performed to preliminarily clean and sort the original data, filter invalid information, and extract information directly useful for a recommendation system, so as to facilitate training of a data model.

Step S905, determining the priority sequence of the recommendation modes based on the field according to the target information, the source domain information, the updated feature information of the target item and the feature information of the source item.

And step S906, determining recommended items by sequentially selecting corresponding recommendation modes according to the priority sequence so as to form a recommendation list corresponding to the terminal.

And step S907, sequentially sending the recommended items to the terminal according to the recommended list.

In other embodiments, after completing the recommendation to the terminal and forming the recommendation list, the method further includes:

step S910, acquiring the recommendation list corresponding to at least part of terminals in the domain-based recommendation system, and adopting a preset recommendation algorithm to form the recommendation list for at least part of terminals in the domain-based recommendation system.

Here, the recommendation list corresponding to at least a part of terminals in the domain-based recommendation system is a recommendation list formed by any one of the domain-based recommendation methods provided in the embodiments of the present application.

The preset recommendation algorithm is a traditional recommendation method, such as any one of a collaborative filtering method, a hybrid recommendation method and the like.

In the embodiment of the application, a part of all terminals in the recommendation system can be recommended by adopting the field-based recommendation method provided by any embodiment of the application to obtain the recommendation list, and the other part of the terminals can be recommended by adopting the traditional recommendation method to obtain the recommendation list, so that each terminal in the recommendation system corresponds to the recommendation list.

And step S911, integrating the acquired recommendation list to form an integral recommendation list.

Here, the obtained recommendation lists of all terminals may be integrated to form the overall recommendation list, and the overall recommendation list integrates information of all terminals in the recommendation system, so that the overall recommendation list is applicable to all terminals, and the overall recommendation list may be used for recommending items to all terminals.

Step S912, performing emotion analysis on the recommended items in the overall recommendation list, and determining emotion orientations of the recommended items.

Here, the emotion analysis may be implemented by: and performing word segmentation processing on the characteristic information of the recommended item, wherein the characteristic information of the recommended item can be brief description information of the recommended item and comment information of all users on the recommended item. And performing feature extraction and feature selection on the feature information after word segmentation processing, generating a classification model according to the selected feature information, and obtaining the emotional orientation of all users to the recommended item through the classification model. Wherein the emotional orientations comprise at least two emotional orientations of a positive side and a negative side, and of course, the emotional orientations can also comprise neutral emotional orientations between the positive side and the negative side.

In the embodiment of the application, after the emotional orientation of the recommended item is determined, the emotional orientation of the recommended item can be added to the feature information of the recommended item to serve as a label of the recommended item, so that the recommended item can be conveniently processed subsequently.

Step S913, adjusting the ranking of the recommended items in the entire recommendation list so that the ranking priority of the recommended item with positive emotion orientation is higher than the ranking priority of the recommended item with negative emotion orientation.

Here, adjusting the ranking of the recommended items in the overall recommendation list can be implemented in the following two ways:

the method I comprises the following steps: and adjusting the recommendation items with positive emotion orientation to the front part of the whole recommendation list.

In this way, in the adjusted overall recommendation list, the recommendation item with the positive emotion orientation is positioned in the front, and the recommendation items with other emotion orientations are positioned behind the recommendation item with the positive emotion orientation, so that the recommendation item with the positive emotion orientation can be recommended to the user first when in recommendation.

It is noted that the other emotional orientations may include negative and neutral emotional orientations. Then, in the adjusted overall recommendation list, the positive recommendation items are located at the front of the overall recommendation list, and the negative and neutral emotional orientations are located behind the positive recommendation items.

Of course, after the positive recommended items are adjusted to the front part of the whole recommended list, the neutral recommended items and the negative recommended items can be adjusted, so that the neutral recommended items are located before the negative recommended items, the neutral recommended items are preferentially recommended after the positive recommended items are recommended, and the negative recommended items are finally recommended.

The second method comprises the following steps: recommended items with negative emotional orientation are deleted.

In this way, the overall recommendation list may be optimized such that negative recommended items that are not suitable for recommendation to the user are not included in the adjusted overall recommendation list.

Of course, after deleting the negative recommended items, if the overall recommendation list includes both the positive recommended items and the neutral recommended items, the positive recommended items may be adjusted to the front of the overall recommendation list, so as to recommend the neutral recommended items to the user after preferentially recommending the positive recommended items to the user.

It should be noted that, in the embodiment of the present application, after the recommendation list of the terminal is determined, emotion analysis may also be performed on the recommendation items in the recommendation list, so as to determine an emotional orientation of each recommendation item in the recommendation list. And adjusting the sequence of the recommended items in the recommended list according to the emotional orientation of each recommended item.

According to the field-based recommendation method provided by the embodiment of the application, after the recommendation lists of the terminals are obtained, the recommendation lists of all the terminals in the recommendation system are integrated to obtain the overall recommendation list, so that a more comprehensive recommendation process for all the terminals is realized. And performing emotion analysis on the recommended items, and adjusting the arrangement sequence of the recommended items in the overall recommendation list or the recommendation list according to the emotion orientation of the recommended items, so as to preferentially recommend forward recommended items to the user, and thus realize personalized recommendation to the user.

Based on the above embodiments, the present application embodiment further provides a field-based recommendation method, and in the first step, context data provided by a mobile terminal (i.e., a terminal) is first obtained to supplement a user portrait, then all APP interaction information of the mobile terminal is collected, mainly for including interaction information of APPs of people who are used by the mobile terminal, a personalized user portrait is obtained, and then a deep learning and task priority-based recommendation method is adopted on a distributed machine learning platform (e.g., a Spark platform) to form a personal recommendation list (i.e., a recommendation list of the terminal), and the personal recommendation list is updated in real time. And secondly, counting data of the personal recommendation lists of all mobile terminals, performing data modeling on a Spark platform by adopting a deep learning method to obtain a more comprehensive classification set, forming an overall recommendation list by adopting a cross-domain recommendation method of task priority or a traditional recommendation method, adding an emotion analysis module, optimizing the personal recommendation list and the overall recommendation list to obtain a final personalized recommendation list, and updating at regular time.

Fig. 10 is a schematic flowchart illustrating an implementation flow of a recommendation method based on task priority according to an embodiment of the present application, where as shown in fig. 10, the method includes the following steps:

step S1001 determines whether the target domain has a similar system at the mobile terminal.

Here, the method of determining whether a similar system exists may be: after obtaining the APP list of the mobile terminal, crawling the software introduction pages of all the APPs in the APP list from the Internet, taking the APP where the item to be recommended is located as a target domain, and calculating the text similarity between the software introduction page of the APP list and the software introduction pages of all other APPs in pairs. The following three methods are generally adopted for calculating the text similarity: a semantic understanding-based calculation method, a vector space model-based calculation method and a hamming distance-based calculation method. And when the similarity of the obtained texts is greater than or equal to 95%, judging the two APPs as similar systems, otherwise, judging the two APPs as non-similar systems.

If it is judged that the similar system exists, step S1002 is executed, and if it is judged that the similar system does not exist, step S1008 is executed.

And step S1002, performing multi-field recommendation by adopting a multi-field recommendation mode.

Step S1003, after the multi-domain recommendation task is completed, judging whether more overlapped information exists between the target domain and the source domain.

Here, the method of determining whether there is more overlapped information may be: firstly, traversing all recommendable items by taking an APP as a target domain, and setting the number of the recommendable items as m; then, the other APPs are the source domains, and all the items are traversed similarly, and the number of the recommendable items in each source domain is set as n. And sequentially calculating the similarity of each item in the target domain and each item in the source domain, wherein the similarity calculation method can adopt cosine similarity, a Pearson correlation coefficient, an Euclidean distance and a Jacard coefficient method, and when the similarity is more than or equal to 95%, the two items are judged to be similar. Let the number of similar items in a certain source domain be t (t ≦ n), when

Then, the target domain and the source domain are determined to have more overlapping information.

If it is determined that there is more overlapping information between the target domain and any one of the source domains, steps S1004 and S1005 are performed, and if it is determined that there is no more overlapping information between the target domain and each of the source domains, steps S1006 and S1007 are performed.

Step S1004, a multi-connection recommendation method is first adopted to perform multi-connection recommendation.

And step S1005, performing cross-domain recommendation by adopting a cross-domain recommendation mode.

And step S1006, performing cross-domain recommendation by adopting a cross-domain recommendation mode.

And step 1007, performing multi-connection recommendation by adopting a multi-connection recommendation mode.

Step S1008, if it is determined that there is no similar system, it is determined whether there is more overlapping information between the target domain and the source domain.

Here, the determination may be made by the same method as the above-described method of determining whether or not there is much overlapped information.

If it is determined that the target domain and any source domain have more overlapping information, step S1009 and step S1010 are performed, and if it is determined that the target domain and each source domain do not have more overlapping information, step S1011 and step S1012 are performed.

In step S1009, a multi-connection recommendation method is first adopted to perform multi-connection recommendation.

And step S1010, performing cross-domain recommendation by adopting a cross-domain recommendation mode.

In step S1011, a cross-domain recommendation method is first adopted to perform cross-domain recommendation.

Step S1012, then, a multi-connection recommendation method is used to perform multi-connection recommendation.

In the embodiment of the present application, the method for performing cross-domain recommendation may be: firstly, a user and project potential factor Matrix is generated by utilizing a comment-oriented Matrix Factorization (MF) model and a ranking-oriented MF model, wherein the comment-oriented MF model comprises the following components: maximum distance Matrix decomposition (MMMF), probability Matrix decomposition (PMF); the ranking-oriented MF model includes: bayesian Personalized Rankings (BPR). And then combining the latent factor matrixes according to the sparsity of each user and each item to generate a reference factor matrix, and training a Deep Neural Network (DNN) according to feedforward and back propagation processes under the guidance of the reference factors to express the relationship of the latent factors between two domains or two systems, so that the latent factors are mapped in the source domain, the rating of the user on the items in the source domain is obtained, and the matched items in the source domain are recommended in the target domain.

In order to highlight the characteristics of a mobile terminal recommendation system, realize personalized real-time recommendation and solve the common problems of low personalization degree, sparse data, low recommendation accuracy and the like in the conventional cross-domain recommendation system at present, the embodiment of the application also adds the specific contextual data of the mobile terminal to supplement a user portrait, collects all APP interaction information of the mobile terminal, mainly aims to contain the interaction information of the APP of the Xiaozhong people used by the mobile terminal, then forms a personal recommendation list on a Spark platform by adopting a deep learning and task priority-based recommendation method, and updates the personal recommendation list in real time. And then counting all mobile terminal recommendation lists, performing data modeling on a Spark platform by adopting a deep learning method to form personalized classification, then forming an overall recommendation list by adopting a task priority-based recommendation method or a traditional recommendation method, finally adding an emotion analysis module, optimizing the individual recommendation list and the overall recommendation list to obtain a final personalized recommendation list, and updating at regular time.

Fig. 11 is a schematic implementation flowchart of a domain-based recommendation method provided in an embodiment of the present application, and as shown in fig. 11, the method includes the following steps:

in step S1101, personal scene data, such as a terminal location, a call frequency, and the like, is acquired by the mobile terminal.

Step S1102, collecting APP lists installed by a single mobile terminal, and obtaining all user interaction information of the mobile terminal to obtain personal APP data, mainly for including a member APP list and interaction information thereof used by the mobile terminal.

Step S1103, preprocessing the collected data, that is, performing preliminary cleaning and sorting on the raw data, filtering out invalid information portions, and extracting information directly useful for the recommendation system.

And step S1104, the preprocessed information is put into a data storage system.

Step S1105, on the Spark platform, feature extraction is carried out on the project of the target domain through the personal recommendation engine, then a recommendation method based on task priority is adopted to generate a personal recommendation list for a single mobile terminal, real-time updating is carried out, and then the personal recommendation list is used as personal APP data, so that the recycling of recommendation information is realized.

Step S1106, counting data of all mobile terminal personal recommendation lists, performing data modeling by using a deep learning method through a comprehensive recommendation engine on a Spark platform to obtain a more comprehensive classification set, and then performing overall recommendation on all mobile terminals by using a task priority-based recommendation method or a conventional recommendation method (e.g., collaborative filtering, hybrid recommendation, etc.), so as to generate an overall recommendation list.

Step S1107, emotion analysis is carried out based on the whole recommendation list, and positive and negative emotion information is separated.

Step S1108, adding the emotion analysis information result obtained in step S1107 to the personal recommendation list of the single mobile terminal generated in step S1105 and the overall recommendation list generated in step S1106, performing recommendation list optimization, generating a final personalized recommendation list, updating the personalized recommendation list at regular time, and using the personalized recommendation list as personal APP data to realize recycling of recommendation information.

The following description will be made in detail by taking the example of recommending a movie by the mobile terminal. When the personal recommendation is generated for the first time, a certain mobile terminal is taken as an example, so that the user is the same person, and the recommendation task can be simplified, as shown in fig. 12A to 12C, which are simplified flow diagrams of a multi-domain recommendation method, a multi-connection recommendation method, and a cross-domain recommendation method provided by the embodiment of the present application. The user at the mobile terminal is U, and the project of the source domain is I _S The item of the target domain is I _T For example, in the multi-domain recommendation method, there is user overlap between the source domain and the target domain, and items in the source domain and the target domain may be recommended to each other according to the source domain data 1201 in the source domain and the target domain data 1202 in the target domain, where a region indicated by a dotted line is an item recommended to a user.

In the embodiment of the application, first, scene data, such as a location where a user is located, a call frequency, and the like, is acquired through a mobile terminal (step S1101); then, acquiring all App information of the mobile terminal through a web crawler, wherein the App information comprises an installed App list and interactive information between a mobile terminal user and an App, and is mainly used for containing a member App list and interactive information thereof used by the mobile terminal and ensuring personalized recommendation, such as App login information, comment information, click rate and the like (step S1102); then, preprocessing the acquired data, such as filtering defective data and performing format conversion on valid data, so as to facilitate training of a data model (step S1103); after all valid data of the mobile terminal are obtained, the valid data are stored, a Hadoop ecosphere can be used for data storage, and a Distributed File System (HDFS) is a Distributed File and is convenient to directly use on a Spark platform (step S1104). After the data collection and data preprocessing work is finished, feature extraction is carried out on the project of the target domain of the mobile terminal on a Spark platform by adopting a deep learning method, and then a recommendation method based on task priority is adopted for movie recommendation, or movie recommendation is carried out aiming at a certain recommendation mode. The APP lists obtained in step S1102 are the target domain and the source domain in the personal recommendation engine, and the generation of the personal recommendation list is based on the mutual information obtained in step S1102 (all information specific to the mobile terminal, including the mutual information on APP of the crowd, there is no omission). The deep learning model can adopt CNN, RNN, MLP and the like.

For example, a recommendation method based on task priority is as follows, taking an migu video as a target domain, firstly searching whether similar systems, such as Aiqiyi, tencent and the like, exist in stored data, if the similar systems exist, completing a multi-domain recommendation mode, aggregating knowledge among the systems, and then performing mutual recommendation by adopting methods based on content, collaborative filtering, mixed recommendation and the like; if the similar system does not exist, searching whether a large amount of information similar to the target domain exists in the stored data, such as bean, cicada, panda reading and the like which are overlapped with the movie, if so, completing a multi-connection recommendation mode, supplementing the movie information in the mikuu video by adopting a knowledge connection and migration method, and then recommending the movie on the mikuu video by adopting methods based on content, collaborative filtering, mixed recommendation and the like; if more overlapped information does not exist, a cross-domain recommendation mode with low recommendation accuracy can be completed firstly, the knowledge in the source domain is migrated to the target domain, for example, if a woman whose user portrait is a woman loving to watch antique movies can be obtained through the migu video, antique interactive information can be found through panning, and related movies are recommended in the migu video. Therefore, a personal recommendation list can be obtained, the personal movie recommendation list is used as personal APP data, recycling of recommendation information is achieved, and meanwhile, the personal mobile terminal recommendation list is updated in real time through a Spark platform (step S1105).

The mobile terminal personal movie recommendation list can be obtained through the steps, in order to make recommendation more comprehensive, all mobile terminal personal recommendation list information needs to be counted, a deep learning method is adopted on a Spark platform for data modeling, a more comprehensive classification set is obtained, the deep learning method can adopt CNN, RNN, MLP and the like, and then a recommendation method based on task priority or a traditional recommendation method (collaborative filtering, mixed recommendation and the like) is adopted for movie recommendation (step S1106). When all APPs of the mobile terminal are recommended, more detailed classification results can be obtained through the deep learning, namely, comprehensive user portrait, such as people who have tags of 'love comedy, love for buying cosmetics, love for strolling academic forum', and the like at the same time, all interaction information among the people can be recommended to each other, personalized recommendation is truly achieved, and interest guidance is obtained.

After the overall movie recommendation list is obtained, emotion analysis can be performed, word segmentation feature extraction and feature selection are performed on comment-type texts to generate a classification model, so that emotion orientation of some movie by all users is obtained, positive and negative emotion distinguishing is performed on the movie (step S1107), movies with positive emotion orientation in the generated recommendation list are placed in the front row, movies with negative emotion orientation in the front row are removed from the recommendation list, and therefore the individual movie recommendation list and the overall movie recommendation list are optimized (step S1108). And obtaining a final personalized movie recommendation list through the steps, updating the recommendation list at regular time, for example, once every 24 hours, and using the recommendation list as personal APP data to realize the cyclic utilization of recommendation information.

The field-based recommendation method provided by the embodiment of the application mainly comprises the following aspects:

on the first hand, the interactive data of a single mobile terminal is obtained, the interactive data comprises the specific scene data of the mobile terminal and all APP interactive information of the mobile terminal, and the personalized recommendation effect is ensured mainly for containing the Xiaozhong APP list and the interactive information thereof used by the mobile terminal.

And in the second aspect, feature extraction is carried out on a single mobile terminal target domain item, then a personal recommendation list is generated (and updated in real time) by adopting a proper recommendation algorithm, and the personal recommendation list is used as personal APP data. The features of all domains of the mobile terminal are integrated to form a personalized user portrait. In the example, a deep learning algorithm with better current feature extraction is adopted on a Spark platform, and then a task priority based recommendation method provided by the application is adopted to generate a personal recommendation list and update the personal recommendation list in real time.

And thirdly, counting the personal recommendation lists of all the mobile terminals, extracting and classifying features to obtain a more comprehensive classification set, and generating an integral recommendation list by adopting a proper recommendation algorithm to realize personalized recommendation and obtain interest guidance. In the example, a deep learning algorithm with good current feature extraction and classification effects is adopted on a Spark platform, and then a recommendation method based on task priority or a traditional recommendation algorithm (collaborative filtering, mixed recommendation and the like) is adopted to generate an overall recommendation list.

And in the fourth aspect, emotion analysis is carried out on the whole recommendation list, positive emotion information and negative emotion information are separated and added into the personal recommendation list and the whole recommendation list for optimization, and finally, a personalized recommendation list is obtained (and is updated regularly) and serves as personal APP data. In the example of the application, word segmentation feature extraction and feature selection are carried out on the comment text to generate a classification model, so that the emotional orientation of all users to a certain item is obtained.

According to the field-based recommendation method provided by the embodiment of the application, the mobile terminal data is fully utilized to supplement the user portrait, and data are independently collected for each mobile terminal, so that the defect that the user cannot obtain personalized recommendation due to the use of a small and popular APP is overcome; in addition, the recommendation modes are prioritized, and the multi-collar recommendation modes which can be completed only based on a similar system are placed at a first priority, so that the recommendation accuracy is improved; and placing the multi-connection recommendation mode which can be completed only when the overlapped domain or the information exists at the second priority, and placing the cross-domain recommendation mode at the last stage as the supplementary recommendation under the conditions of interest supplement and data sparseness.

Based on the foregoing embodiments, the present application provides a recommendation device based on a domain, where the device includes modules and components included in the modules, and may be implemented by a processor in a recommendation device based on a domain; of course, it may also be implemented by logic circuitry; in implementation, the processor may be a Central Processing Unit (CPU), a Microprocessor (MPU), a Digital Signal Processor (DSP), a Field Programmable Gate Array (FPGA), or the like.

Fig. 13 is a schematic structural diagram of a component of a domain-based recommendation apparatus according to an embodiment of the present application, and as shown in fig. 13, the domain-based recommendation apparatus 1300 includes:

an obtaining unit 1301, configured to obtain target information of a target domain on a terminal, source domain information of all source domains, feature information of a target item in the target domain, and feature information of source items in all source domains;

a determining unit 1302, configured to determine, according to the target information, the source domain information, the feature information of the target item, and the feature information of the source item, a priority ranking of a recommendation mode based on a domain;

the processing unit 1303 is configured to select corresponding recommendation manners in sequence to determine recommendation items according to the priority ranking, so as to form a recommendation list corresponding to the terminal;

a sending unit 1304, configured to send the recommended items to a terminal in sequence according to the recommendation list.

In other embodiments, the determining unit includes: the first determining module is used for determining that the multi-domain recommendation mode has a first priority when the first similarity between the target information and the source domain information of any source domain is greater than or equal to a first preset threshold; the second determining module is used for determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority when the number of the first source items in any source domain meets a preset condition; the second similarity between the feature information of the first source item and the feature information of the target item is greater than or equal to a second preset threshold; the first priority is higher than the second priority; the second priority is higher than the third priority.

In other embodiments, the determining unit includes: a third determining module, configured to determine that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority when the first similarity between the target information and the source domain information of each source domain is smaller than a first preset threshold and the number of the first source items in any source domain meets a preset condition; and the second similarity between the characteristic information of the first source item and the characteristic information of the target item is greater than or equal to a second preset threshold value.

In other embodiments, the apparatus further comprises: a second determining unit, configured to determine that the cross-domain recommendation manner has a second priority and the multi-connection recommendation manner has a third priority when the number of the first source items in all the source domains does not satisfy the preset condition.

In other embodiments, the processing unit comprises: the first processing module is used for determining items in a source domain of a user which is the same as the target domain as the recommended items when the selected recommendation mode is a multi-domain recommendation mode so as to form a recommendation list corresponding to the terminal; the second processing module is used for adding the source domain information of the source domain to the target information of the target domain when the selected recommendation mode is a multi-connection recommendation mode, so as to determine recommendation items according to the added source domain information and form a recommendation list corresponding to the terminal; and the third processing module is used for determining the items in the source domains which belong to different domains from the target domain as the recommended items to form a recommendation list corresponding to the terminal when the selected recommendation mode is a cross-domain recommendation mode.

In other embodiments, the apparatus further comprises: a second obtaining unit, configured to obtain scene data of the terminal; and the updating unit is used for updating the characteristic information of the target item in the target domain according to the scene data.

In other embodiments, the apparatus further comprises: and the preprocessing unit is used for preprocessing the target information of the target domain and the source domain information of all the source domains so as to filter out invalid information in the target information and the source domain information.

In other embodiments, the apparatus further comprises: the third acquisition unit is used for acquiring the recommendation lists corresponding to at least part of terminals in the domain-based recommendation system and recommending lists formed by adopting a preset recommendation algorithm on at least part of terminals in the domain-based recommendation system; and the integration unit is used for integrating the acquired recommendation list to form an integral recommendation list.

In other embodiments, the apparatus further comprises: the emotion analysis unit is used for carrying out emotion analysis on recommended items in the whole recommendation list and determining emotion orientations of the recommended items; and the adjusting unit is used for adjusting the sequencing of the recommended items in the whole recommendation list so that the sequencing priority of the recommended items with positive emotion orientation is higher than that of the recommended items with negative emotion orientation.

It should be noted that, in the embodiment of the present application, if the domain-based recommendation method is implemented in the form of a software functional module and is sold or used as a standalone product, it may also be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application or portions thereof that contribute to the related art may be embodied in the form of a software product, where the computer software product is stored in a storage medium and includes several instructions for enabling a terminal to execute all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read Only Memory (ROM), a magnetic disk, or an optical disk. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.

Correspondingly, an embodiment of the present application provides a recommendation device based on a domain, and fig. 14 is a schematic structural diagram of the recommendation device based on a domain provided in the embodiment of the present application, as shown in fig. 14, the recommendation device 1400 based on a domain at least includes: a processor 1401, a communication interface 1402, and a storage medium 1403 configured to store executable instructions, wherein the processor 1401 generally controls the overall operation of the recommendation device.

The communication interface 1402 may enable the recommendation device to communicate with other terminals or servers via a network.

The storage medium 1403 is configured to store instructions and applications executable by the processor 1401, and may also cache data to be processed or processed by each module in the processor 1401 and the recommendation device 1400, and may be implemented by a FLASH Memory (FLASH) or a Random Access Memory (RAM).

It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in the various embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.

It should be noted that, in this document, the terms "comprises", "comprising" or any other variation thereof are intended to cover a non-exclusive inclusion, so that a process, a method or an apparatus including a series of elements includes not only those elements but also other elements not explicitly listed or inherent to such a process, a method or an apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element. In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above-described device embodiments are merely illustrative, for example, the division of the unit is only one logical function division, and there may be other division ways in actual implementation, such as: multiple units or components may be combined, or may be integrated into another system, or some features may be omitted, or not implemented.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units; can be located in one place or distributed on a plurality of network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of the present application. Those of ordinary skill in the art will understand that: all or part of the steps for realizing the method embodiments can be completed by hardware related to program instructions, the program can be stored in a computer readable storage medium, and the program executes the steps comprising the method embodiments when executed; and the aforementioned storage medium includes: various media that can store program code, such as removable storage devices, read-only memories, magnetic or optical disks, etc. Alternatively, the integrated units described above in the present application may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as independent products. Based on such understanding, the technical solutions of the embodiments of the present application may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a terminal to execute all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a removable storage device, a ROM, a magnetic or optical disk, or other various media that can store program code.

The above description is only an embodiment of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think of the changes or substitutions within the technical scope of the present application, and shall cover the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (9)

1. A domain-based recommendation method, the method comprising:

acquiring target information of a target domain, source domain information of all source domains, characteristic information of target items in the target domain and characteristic information of source items in all source domains on a terminal;

when the first similarity between the target information and the source domain information of any source domain is larger than or equal to a first preset threshold value, determining that the multi-domain recommendation mode has a first priority; the multi-domain recommendation mode is a mode of recommending the items of the source domain and the target domain to the user in the source domain and the user in the target domain;

when the number of the first source items in any source domain meets a preset condition, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority; or when the first similarity between the target information and the source domain information of each source domain is smaller than a first preset threshold and the number of the first source items in any source domain meets a preset condition, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority; the multi-connection recommendation mode is a mode of enriching the available knowledge of the target domain through the source domain information collected by the source domain; the cross-domain recommendation mode is a mode of recommending items of the target domain to the user of the source domain through the knowledge of the target domain;

when the number of the first source items in all the source domains does not meet the preset condition, determining that the cross-domain recommendation mode has a second priority, and the multi-connection recommendation mode has a third priority;

when the recommendation mode selected according to the priority ranking is a multi-field recommendation mode, determining items in a source domain having the same user as the target domain as recommendation items to form a recommendation list corresponding to the terminal;

when the recommendation mode selected according to the priority ranking is a multi-connection recommendation mode, adding the source domain information of the source domain to the target information of the target domain to determine the recommendation items according to the added source domain information to form a recommendation list corresponding to the terminal;

when the recommendation mode selected according to the priority ranking is a cross-domain recommendation mode, determining items in source domains which belong to different domains from the target domain as the recommendation items to form a recommendation list corresponding to the terminal;

and sequentially sending the recommended items to the terminal according to the recommended list.

2. The method according to claim 1, wherein a second similarity between the feature information of the first source item and the feature information of the target item is greater than or equal to a second preset threshold;

the first priority is higher than the second priority; the second priority is higher than the third priority.

3. The method of claim 1, further comprising:

before acquiring target information of a target domain, source domain information of all source domains, characteristic information of a target item in the target domain and characteristic information of source items in all source domains on a terminal, acquiring scene data of the terminal;

and updating the characteristic information of the target item in the target domain according to the scene data.

4. The method of claim 1, further comprising:

and preprocessing the target information of the target domain and the source domain information of all the source domains to filter out invalid information in the target information and the source domain information.

5. The method of claim 1, further comprising:

acquiring the recommendation lists corresponding to at least part of terminals in the field-based recommendation system, and adopting a preset recommendation algorithm to form the recommendation lists for at least part of terminals in the field-based recommendation system;

and integrating the acquired recommendation lists to form an integral recommendation list.

6. The method of claim 5, further comprising:

performing emotion analysis on recommended items in the overall recommendation list, and determining emotion orientations of the recommended items;

and adjusting the sequence of the recommended items in the whole recommendation list so that the sequence priority of the recommended items with positive emotion orientation is higher than the sequence priority of the recommended items with negative emotion orientation.

7. A domain-based recommendation apparatus, the apparatus comprising:

an acquiring unit, configured to acquire target information of a target domain on a terminal, source domain information of all source domains, feature information of a target item in the target domain, and feature information of source items in all source domains;

the determining unit is used for determining that the multi-domain recommendation mode has a first priority when the first similarity between the target information and the source domain information of any source domain is greater than or equal to a first preset threshold; the multi-domain recommendation mode is a mode of recommending the items of the source domain and the target domain to the user in the source domain and the user in the target domain; when the number of the first source items in any source domain meets a preset condition, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority; or when the first similarity between the target information and the source domain information of each source domain is smaller than a first preset threshold and the number of the first source items in any source domain meets a preset condition, determining that the multi-connection recommendation mode has a second priority and the cross-domain recommendation mode has a third priority; the multi-connection recommendation mode is a mode of enriching the available knowledge of the target domain through the source domain information collected by the source domain; the cross-domain recommendation mode is a mode of recommending items of the target domain to the user of the source domain through the knowledge of the target domain; when the number of the first source items in all the source domains does not meet the preset condition, determining that the cross-domain recommendation mode has a second priority, and the multi-connection recommendation mode has a third priority;

the processing unit is used for determining items in a source domain with the same user as the target domain as recommended items to form a recommendation list corresponding to the terminal when the recommendation mode selected according to the priority ranking is a multi-domain recommendation mode; when the recommendation mode selected according to the priority ranking is a multi-connection recommendation mode, adding the source domain information of the source domain to the target information of the target domain to determine the recommendation items according to the added source domain information to form a recommendation list corresponding to the terminal; when the recommendation mode selected according to the priority ranking is a cross-domain recommendation mode, determining items in source domains which belong to different domains from the target domain as the recommendation items to form a recommendation list corresponding to the terminal;

and the sending unit is used for sequentially sending the recommended items to the terminal according to the recommended list.

8. A domain-based recommendation device, characterized in that the device comprises at least: a processor and a storage medium configured to store executable instructions, wherein: the processor is configured to execute stored executable instructions;

the executable instructions are configured to perform the domain-based recommendation method provided by any of the preceding claims 1 to 6.

9. A storage medium having stored thereon computer-executable instructions configured to perform the domain-based recommendation method provided by any of claims 1 to 6.

Images