CN113032588A

CN113032588A - Multimedia recommendation system, method, device and equipment

Info

Publication number: CN113032588A
Application number: CN201911357766.4A
Authority: CN
Inventors: 牛亚男; 冷德维
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2019-12-25
Filing date: 2019-12-25
Publication date: 2021-06-25
Anticipated expiration: 2039-12-25
Also published as: CN113032588B

Abstract

The disclosure provides a multimedia recommendation system, method, device and equipment, and belongs to the technical field of internet. The system comprises: a recommendation device and a feature service device; the recommendation device is used for acquiring the user characteristics corresponding to the user information in the recommendation processing request and sending the information identification of the multimedia information to the characteristic service device; the characteristic service device is used for reading the multimedia characteristics corresponding to the information identification; and the recommending device is also used for splicing the user characteristics and the multimedia characteristics and acquiring the matching degree of the user information and each multimedia information according to the obtained splicing characteristics. Compared with the mode that one device is used for carrying out feature storage, feature reading and matching degree obtaining, the efficiency of information recommendation of the multimedia recommendation system is improved.

Description

Multimedia recommendation system, method, device and equipment

Technical Field

The present disclosure relates to the field of internet technologies, and in particular, to a multimedia recommendation system, method, apparatus, and device.

Background

Nowadays, with the development of internet technology, entertainment applications on terminals are increasing. For example, a short video application is installed on the terminal, and a user can watch a short video through the short video application. To increase the exposure rate of the short video, the short video application may also recommend to the user some short videos that may be of interest to the user. When the short video application is used for recommending short videos for a user, the probability of interest of the user to a plurality of short videos to be recommended can be determined by means of a multimedia recommendation system, and the short videos which are interested by the user are recommended for the user according to the probability of interest of the user to each short video.

Disclosure of Invention

The embodiment of the disclosure provides a multimedia recommendation system, a method, a device and equipment, which can solve the problem of low information recommendation efficiency of the multimedia recommendation system in the related art. The technical scheme is as follows:

in a first aspect, a multimedia recommendation system is provided, which includes: a recommendation device and a feature service device;

the recommendation device is used for acquiring the user characteristics corresponding to the user information and sending the information identification of the multimedia information to the characteristic service device after receiving a recommendation processing request carrying the user information and a plurality of multimedia information;

the feature service device is used for reading the multimedia features corresponding to the information identifications of the multimedia information and returning the read multimedia features to the recommendation device;

and the recommendation device is also used for splicing the user characteristics and the read multimedia characteristics to obtain splicing characteristics, and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics.

In one possible implementation manner, the recommendation device includes M recommendation units, where M is an integer greater than 1;

each recommending unit is used for acquiring the user characteristics of the corresponding user information and sending the information identification of part of the multimedia information in the plurality of multimedia information to the characteristic service device, and no intersection exists between the part of the multimedia information corresponding to any two recommending units;

each recommending unit is also used for splicing the user characteristics with the multimedia characteristics of the corresponding multimedia information returned by the characteristic service device to obtain splicing characteristics; determining the matching degree of the user information and the corresponding multimedia information according to the splicing characteristics;

and the recommending device is also used for merging the matching degree of the user information obtained by each recommending unit and the corresponding multimedia information and returning the merged user information to a sender of the recommending processing request.

In another possible implementation manner, the feature service device includes N feature service units, where N is an integer greater than 1;

each feature service unit is used for storing part of multimedia features, and no intersection exists between the part of the multimedia features stored by any two feature service units;

the feature service unit is further configured to read a multimedia feature corresponding to the information identifier of the received multimedia information from a part of the stored multimedia features, and return the read multimedia feature to the corresponding recommendation unit.

In another possible implementation manner, the recommendation apparatus further includes a request processing unit;

the request processing unit is configured to receive the recommendation processing request, allocate the user information and the plurality of multimedia information in the recommendation processing request to the M recommendation units, and no intersection exists between the multimedia information allocated by any two recommendation units.

In another possible implementation manner, each recommending unit is further configured to send the matching degree between the user information and each multimedia information in the part of multimedia information to the request processing unit;

the request processing unit is further configured to sort the plurality of multimedia information according to the matching degree between the user information and each multimedia information in the plurality of media information, and output the sorted plurality of multimedia information.

In another possible implementation manner, the multimedia recommendation system further includes: an offline feature training device;

the offline feature training device is used for training multimedia features and synchronizing the multimedia features to the N feature service units.

In a possible implementation manner, the offline feature training device is further configured to determine, when a multimedia feature is added, a feature service unit in which the multimedia feature is stored, and store the multimedia feature to a corresponding feature service unit.

In a second aspect, a multimedia recommendation method is provided, where the method is applied in a recommendation device, and the method includes:

receiving a recommendation processing request carrying user information and a plurality of multimedia information, acquiring user characteristics corresponding to the user information, and sending information identification of the multimedia information to a characteristic service device;

and splicing the user characteristics and the multimedia characteristics read by the characteristic service device to obtain splicing characteristics, and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics.

the sending of the information identifier of the multimedia information to the feature service device includes:

distributing the plurality of multimedia information to the M recommending units, wherein no intersection exists between the multimedia information distributed by any two recommending units, and respectively sending information identifiers of the multimedia information to the feature service device through the M recommending units;

the splicing of the user characteristics and the multimedia characteristics read by the characteristic service device to obtain splicing characteristics, and obtaining the matching degree of the user information and each multimedia information according to the splicing characteristics comprise:

splicing the user characteristics and the multimedia characteristics of the corresponding multimedia information returned by the characteristic service device through the M recommending units to obtain spliced characteristics; and acquiring the matching degree of the user information and the corresponding multimedia information according to the splicing characteristics.

In another possible implementation manner, the obtaining the matching degree between the user information and each multimedia information according to the splicing feature includes:

and inputting the splicing characteristics into a probability prediction model to obtain the matching degree of the user information and each multimedia information.

In another possible implementation manner, the allocating the plurality of multimedia information to the M recommending units includes:

averagely distributing the plurality of multimedia information to the M recommending units; or distributing the plurality of multimedia information to the M recommending units according to the load of each recommending unit and the load balancing principle.

In another possible implementation manner, after obtaining the matching degree between the user information and each multimedia information, the method further includes:

and sequencing the plurality of multimedia information according to the matching degree of the user information and each multimedia information, and outputting the sequenced plurality of multimedia information.

In a third aspect, another multimedia recommendation method is provided, which is applied in a feature service device, and the method includes:

receiving information identification of each multimedia information sent by a recommending device;

reading the multimedia characteristics corresponding to the information identification of each multimedia information according to the information identification of each multimedia information;

and returning the read multimedia features to the recommending device.

In one possible implementation manner, the feature service device includes N feature service units, where N is an integer greater than 1; each feature service unit stores part of multimedia features, and intersection does not exist between the part of the multimedia features stored by any two feature service units;

the reading of the multimedia features corresponding to the information identifier of each multimedia information according to the information identifier of each multimedia information includes:

respectively reading multimedia characteristics corresponding to the information identification of the multimedia information from part of the multimedia characteristics stored by the N characteristic service units;

the returning of the read multimedia features to the recommendation device includes:

and respectively returning the read multimedia features to the recommendation device through the N feature service units.

In a fourth aspect, another multimedia recommendation method is provided, which is applied in an offline feature training apparatus, and includes:

training multimedia features, and synchronizing the multimedia features into a feature service device; the feature service device comprises N feature service units, wherein N is an integer greater than 1; each feature service unit stores part of multimedia features, and intersection does not exist between the part of the multimedia features stored by any two feature service units;

when a multimedia feature is added, determining a feature service unit in which the multimedia feature is stored;

and storing the multimedia characteristics to the determined characteristic service unit.

In a possible implementation manner, the determining, when a multimedia feature is added, a feature service unit in which the multimedia feature is stored includes:

when multimedia features are added, carrying out hash operation on the multimedia features to obtain hash values of the multimedia features; and mapping the N according to the hash value to obtain a feature service unit in which the multimedia features are stored.

In a fifth aspect, a recommendation device is provided, the device comprising:

the recommendation processing request receiving module is configured to execute a recommendation processing request carrying user information and a plurality of multimedia information;

the user characteristic acquisition module is configured to execute user characteristic acquisition corresponding to the user information;

an information identifier sending module configured to execute sending of an information identifier of the multimedia information to the feature service device;

and the matching degree acquisition module is configured to splice the user characteristics and the multimedia characteristics read by the characteristic service device to obtain splicing characteristics, and acquire the matching degree between the user information and each piece of multimedia information according to the splicing characteristics.

the information identifier sending module is further configured to perform assigning the plurality of multimedia information to the M recommending units, and send the information identifiers of the multimedia information to the feature service device through the M recommending units respectively, where no intersection exists between the multimedia information assigned by any two recommending units;

the matching degree obtaining module is configured to perform splicing on the user characteristics and the multimedia characteristics of the corresponding multimedia information returned by the characteristic service device through the M recommending units to obtain spliced characteristics; and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics.

In another possible implementation manner, the matching degree obtaining module is configured to input the splicing features into a probability prediction model to obtain the matching degree between the user information and each multimedia information.

In another possible implementation manner, the information identifier sending module is configured to perform an average assignment of the plurality of multimedia information to the M recommending units; or distributing the plurality of multimedia information to the M recommending units according to the load of each recommending unit and the load balancing principle.

In another possible implementation manner, the apparatus further includes:

and the multimedia information output module is configured to sort the plurality of multimedia information according to the matching degree of the user information and each multimedia information, and output the sorted plurality of multimedia information.

In a sixth aspect, there is provided a feature service apparatus, the apparatus comprising:

the information identifier receiving module is configured to execute receiving of the information identifier of each piece of multimedia information sent by the recommending device;

the multimedia characteristic reading module is configured to execute reading of multimedia characteristics corresponding to the information identification of each multimedia information according to the information identification of each multimedia information;

and the multimedia characteristic sending module is configured to return the read multimedia characteristics to the recommending device.

the multimedia feature reading module is configured to perform reading of multimedia features corresponding to information identifications of the multimedia information from part of the multimedia features stored by the N feature service units respectively;

the multimedia feature sending module is configured to execute returning of the read multimedia features to the recommending device through the N feature service units respectively.

In a seventh aspect, there is provided an offline feature training device, comprising:

a feature training module configured to perform training multimedia features, synchronizing the multimedia features into a feature service device; the feature service device comprises N feature service units, wherein N is an integer greater than 1; each feature service unit stores part of multimedia features, and intersection does not exist between the part of the multimedia features stored by any two feature service units;

a feature service unit determination module configured to perform, when a multimedia feature is added, determining a feature service unit in which the multimedia feature is stored;

and the feature storage module is configured to store the multimedia features into the determined feature service unit.

In a possible implementation manner, the feature service unit determining module is configured to perform a hash operation on a multimedia feature when the multimedia feature is added, so as to obtain a hash value of the multimedia feature; and mapping the N according to the hash value to obtain a feature service unit in which the multimedia features are stored.

In an eighth aspect, a multimedia recommendation apparatus is provided, the apparatus comprising: one or more processors; volatile or non-volatile memory for storing the one or more processor-executable instructions; wherein the one or more processors are configured to perform any of the multimedia recommendation methods described above.

In a ninth aspect, a non-transitory computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions, and the instructions, when executed by a processor of a server, implement any of the multimedia recommendation methods described above.

The technical scheme provided by the embodiment of the disclosure has the following beneficial effects:

in an embodiment of the present disclosure, a multimedia recommendation system includes: a recommendation device and a feature service device; the recommendation device is used for acquiring the user characteristics corresponding to the user information and sending the information identification of the multimedia information to the characteristic service device after receiving a recommendation processing request carrying the user information and a plurality of multimedia information; the feature service device is used for reading the multimedia features corresponding to the information identifications of the multimedia information and returning the read multimedia features to the recommendation device; and the recommendation device is also used for splicing the user characteristics and the read multimedia characteristics to obtain splicing characteristics, and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics. Because the multimedia recommendation system comprises the recommendation device and the feature service device, the recommendation device is responsible for acquiring the information identification of the multimedia information and the matching degree of the user information and each piece of multimedia information, and the feature service device is responsible for storing the features and reading the features according to the information identification, compared with the method that the features are stored, read and acquire the matching degree through one device, the efficiency of information recommendation of the multimedia recommendation system is improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.

FIG. 1 is a schematic illustration of an implementation environment provided by embodiments of the present disclosure;

FIG. 2 is a schematic diagram of a multimedia recommendation system provided by an embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a multimedia recommendation system provided by an embodiment of the present disclosure;

FIG. 4 is a schematic diagram of a multimedia recommendation system provided by an embodiment of the present disclosure;

FIG. 5 is a schematic diagram of a multimedia recommendation system provided by an embodiment of the present disclosure;

fig. 6 is a flowchart of a multimedia recommendation method provided by an embodiment of the present disclosure;

fig. 7 is a flowchart of a multimedia recommendation method provided by an embodiment of the present disclosure;

fig. 8 is a flowchart of a multimedia recommendation method provided by an embodiment of the present disclosure;

fig. 9 is a block diagram of a recommendation device provided by an embodiment of the present disclosure;

FIG. 10 is a block diagram of a feature service apparatus provided by an embodiment of the present disclosure;

FIG. 11 is a block diagram of an offline feature training apparatus provided by an embodiment of the present disclosure;

fig. 12 is a schematic structural diagram of a multimedia recommendation device provided in an embodiment of the present disclosure.

The reference numerals denote:

10-terminal, 20-multimedia server, 30-multimedia recommendation system, 31-recommendation device, 32-feature service device, 33-off-line feature training device, 311-recommendation unit, 312-request processing unit and 321-feature service unit.

Detailed Description

To make the objects, technical solutions and advantages of the present disclosure more apparent, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

Fig. 1 is a schematic diagram of an application scenario provided in an embodiment of the present disclosure. Referring to fig. 1, the application scenario includes a terminal 10, a multimedia server 20, and a multimedia recommendation system 30. Wherein the terminal 10 and the multimedia server 20 are connected through a wireless or wired network. And, a multimedia application for providing a multimedia service is installed on the terminal 10, and a corresponding user of the terminal 10 can implement functions such as data transmission, message interaction, etc. through the multimedia application. The multimedia server 20 is connected to the multimedia recommendation system 30. The terminal 10 may be a computer, a mobile phone, a tablet computer, or other electronic devices. The multimedia application may be a short video application or an audio playback application, etc.

When the terminal 10 logs in to the multimedia server 20 through the multimedia application, the multimedia server 20 can recommend multimedia information to the terminal 10 by means of the multimedia recommendation system 30. Correspondingly, the multimedia server 20 is configured to send a recommendation processing request to the multimedia recommendation system 30, where the recommendation processing request carries the multimedia information of the user. The multimedia information includes at least a plurality of multimedia information and may further include user information.

The user information may include at least one of a gender, an age, a living area, an occupation, a school, a specialty, a school calendar, a favorite subject, a favorite food, a favorite star, a favorite makeup blogger, a favorite movie type, a favorite art, a favorite song type, a favorite animal, a favorite tourist attraction, a favorite brand, a favorite sports item, and the like of the user.

The multimedia information may include at least one of video information and audio information. The video information may include short video. The audio information may include at least one of a song, a talking novel, a comment, a phase sound, etc., and the present disclosure is described with multimedia information as an example of short video.

The multimedia recommendation system 30 receives the recommendation processing request sent by the multimedia server 20, and obtains the matching degree between the user information and each multimedia information according to the user information and the multimedia information. In one possible implementation, the multimedia recommendation system 30 sends the matching degree of the user information and each multimedia information to the multimedia server 20. The multimedia server 20 recommends multimedia information for the user according to the degree of matching between the user information and each multimedia information.

In another possible implementation manner, the multimedia recommendation system 30 sorts the plurality of multimedia information according to the matching degree between the user information and each multimedia information, and sends the sorted plurality of multimedia information to the multimedia server 20. The multimedia server 20 recommends multimedia information for the user according to each of the sorted multimedia information.

For example, when the multimedia application is a short video application, the multimedia server 20 is a short video application server accordingly. When the user watches a short video through the short video application on the terminal 10, the short video application server transmits a recommendation processing request to the multimedia recommendation system 30. The multimedia recommendation system 30 obtains the matching degree of the user information and each short video according to the user information and the short videos in the recommendation processing request, and the multimedia recommendation system 30 can also sequence the short videos according to the matching degree of the user information and each short video, and send the sequenced short videos to the short video application server. After receiving the sequenced short videos, the short video application server sequentially recommends the sequenced short videos to the terminal 10.

Fig. 2 is a schematic diagram of a multimedia recommendation system provided in an embodiment of the present disclosure. Referring to fig. 2, the multimedia recommendation system includes: the recommendation device is connected with the feature service device;

the recommendation device is used for acquiring user characteristics corresponding to the user information and sending information identification of the multimedia information to the characteristic service device after receiving a recommendation processing request carrying the user information and a plurality of multimedia information;

the characteristic service device is used for reading the multimedia characteristics corresponding to the information identification of each piece of multimedia information and returning the read multimedia characteristics to the recommendation device;

and the recommending device is also used for splicing the user characteristics and the read multimedia characteristics to obtain splicing characteristics, and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics.

In an embodiment of the present disclosure, a multimedia recommendation system includes: a recommendation device and a feature service device; the recommendation device is used for acquiring user characteristics corresponding to the user information and sending information identification of the multimedia information to the characteristic service device after receiving a recommendation processing request carrying the user information and a plurality of multimedia information; the characteristic service device is used for reading the multimedia characteristics corresponding to the information identification of each piece of multimedia information and returning the read multimedia characteristics to the recommendation device; and the recommending device is also used for splicing the user characteristics and the read multimedia characteristics to obtain splicing characteristics, and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics. Because the multimedia recommendation system comprises the recommendation device and the feature service device, the recommendation device is responsible for acquiring the information identification of the multimedia information and the matching degree of the user information and each piece of multimedia information, and the feature service device is responsible for storing the features and reading the features according to the information identification, compared with the method that the features are stored, read and acquire the matching degree through one device, the efficiency of information recommendation of the multimedia recommendation system is improved.

In a possible implementation manner, the manner in which the recommendation device obtains the user characteristics corresponding to the user information may be: the recommending means abstracts the user characteristics of the user from the user information.

Wherein, when the user information includes at least one of the gender, age, living area, occupation, school, specialty, academic calendar, favorite subject, favorite food, favorite star, favorite makeup blogger, favorite movie type, favorite general skill, favorite song type, favorite animal, favorite tourist attraction, favorite brand, favorite sports item, etc. of the user, the user characteristic corresponding to the user information includes at least one of the gender characteristic, age characteristic, area characteristic, occupation characteristic, school characteristic, professional characteristic, academic characteristic, food characteristic, star characteristic, makeup blogger characteristic, movie type characteristic, general skill characteristic, song characteristic, animal characteristic, tourist attraction characteristic, brand characteristic, sports item characteristic, etc. of the user.

The recommending device is also used for sending the user characteristics of the user to the characteristic service device.

The recommending device is further used for acquiring the information identifier of each piece of multimedia information according to the plurality of pieces of multimedia information. The implementation mode can be as follows: the recommending device abstracts the information identification of the multimedia information from each piece of multimedia information. The information identifier of the multimedia information can comprise one or more of a gender information identifier, an age information identifier, an area information identifier, an occupation information identifier, a school information identifier, a professional information identifier, a academic information identifier, a subject information identifier, a food information identifier, a star information identifier, a makeup and play blogger information identifier, a movie type information identifier, a comprehensive art information identifier, a song information identifier, an animal information identifier, a tourist attraction information identifier, a brand information identifier, a sports item information identifier and the like. Of course, the information identifier may also include other information identifiers, such as a building information identifier, a cultural information identifier, and the like, and the content and the number of the information identifiers of the multimedia information are not limited by the present disclosure.

For example, when the multimedia information is a short video, and the content of the short video is a video clip of a certain sports star participating in a certain art, the manner of acquiring the information identifier of the multimedia information by the recommending device according to the multimedia information may be: the recommendation device acquires the gender information identification of the sports star in the short video: "male", age information identification: "age 45", area information identification: "China", occupational information identification: table tennis coach, professional information identification: table tennis, and the information identification of the study calendar: "university", star information identification: zhangthree, and the synthesis information identification: "sports star" and the like, the "man", "45 years", "china", "table tennis trainer", "table tennis", "university", "zhang san", "sports star" are used as the information identifier of the short video.

The feature service device is further used for receiving the information identifier of each piece of multimedia information sent by the recommending device and reading the multimedia features corresponding to the information identifiers of the pieces of multimedia information.

The feature service device is further configured to store the multimedia features, that is, to store a corresponding relationship between the information identifier and the feature value of each multimedia feature. Correspondingly, the way for the feature service device to read the multimedia features corresponding to the information identifiers of the multimedia information is as follows: and the characteristic service device reads the characteristic value corresponding to each information identifier from the corresponding relation between the information identifier and the characteristic value according to the information identifier of each multimedia information.

The characteristic service device is also used for storing the corresponding relation between the user characteristics and the characteristic values. The characteristic service device is also used for receiving the user characteristics sent by the recommending device and reading the characteristic values corresponding to the user characteristics from the corresponding relation between the user characteristics and the characteristic values according to the received user characteristics.

In the embodiment of the disclosure, since the multimedia recommendation system includes the recommendation device and the feature service device, the feature service device is responsible for storing the feature and returning the feature according to the information identifier, that is, the feature service device is used as an independent module for storing and returning the feature value in the multimedia recommendation system, so that the feature service device can be used not only by the multimedia recommendation system, but also by other multimedia recommendation systems. In addition, the recommendation device does not need to store the features and read the feature values, so that the space of the recommendation device is saved, and compared with the method that the features are read through one module and the matching degree of the acquired information is obtained, 2/3 memory is saved by the recommendation device, and the recommendation device is convenient to optimize.

In a possible implementation manner, the feature service device is configured to, when the feature value of the user feature and the feature value of the multimedia feature of each piece of multimedia information are read, return the read feature value of the user feature and the read feature value of the multimedia feature of each piece of multimedia information to the recommendation device. The recommending device is used for splicing the characteristic value of the user characteristic with the characteristic value of the multimedia characteristic of each piece of multimedia information to obtain a splicing characteristic, and determining the matching degree of the user information and each piece of multimedia information according to the splicing characteristic.

The recommendation device comprises a probability prediction model, and correspondingly, the mode for determining the matching degree of the user information and each piece of multimedia information according to the splicing characteristics can be as follows: and the recommending device inputs the splicing characteristics into the probability prediction model to obtain the probability value of the matching between the user information output by the probability prediction model and each piece of multimedia information, and the corresponding probability value is used as the matching degree between the user information and the corresponding multimedia information.

The probability prediction model may be a DNN (Deep Neural network) model, a CNN (Convolutional Neural network) model, an RNN (Recurrent Neural network) model, or other Neural network models, and the disclosure does not limit which model is specifically adopted as the probability prediction model.

And the recommending device is also used for outputting the matching degree of the user information and each piece of multimedia information to the multimedia server. Therefore, the multimedia server can recommend the multimedia information according to the matching degree of the user information and each piece of multimedia information. When the multimedia information is the short video, the multimedia server can recommend the short video to the user according to the matching degree of the user information and each multimedia information, so that the browsing amount of the user is increased, and the exposure of the short video is increased.

In a possible implementation manner, the multimedia recommendation system further includes an offline feature training device, and the offline feature training device is connected with the feature service device. And the off-line characteristic training device is used for training the multimedia characteristics and synchronizing the multimedia characteristics to the characteristic service device.

Fig. 3 is a schematic diagram of a multimedia recommendation system provided in an embodiment of the present disclosure. Referring to fig. 3, the recommendation device includes M recommendation units, M being an integer greater than 1;

each recommending unit is connected with the characteristic service device;

each recommending unit is used for acquiring the user characteristics of the corresponding user information and the information identification of part of the multimedia information in the plurality of multimedia information; and sending information identification of part of the multimedia information in the plurality of multimedia information to the feature service device, wherein no intersection exists between the part of the multimedia information corresponding to any two recommendation units.

Each recommending unit is also used for splicing the user characteristics with the multimedia characteristics of the corresponding multimedia information returned by the characteristic service device to obtain splicing characteristics; and determining the matching degree of the user information and the corresponding multimedia information according to the splicing characteristics.

The recommending device is also used for merging the matching degree of the user information obtained by each recommending unit and the corresponding multimedia information and returning the merged user information and the corresponding multimedia information to a sender of the recommending processing request.

Each recommending unit is also used for sending the user characteristics of the corresponding user information to the characteristic service device.

The number of recommendation units may be configured as needed, and in the embodiment of the present disclosure, the number of recommendation units is illustrated as 8. For example, with reference to FIG. 3, the number of recommendation units is 8, recommendation unit 0, recommendation unit 1,. cndot. cndot.7, respectively. The recommending unit is an offer board obtained by slicing the recommending device.

The number of the multimedia information processed by each recommending unit can be the same or different; also, the amount of multimedia information processed by each recommending unit can be allocated by a separate request processing unit. Correspondingly, referring to fig. 4, the recommendation device further includes a request processing unit; the request processing unit is connected with each recommending unit; the request processing unit is used for receiving the recommendation processing request, distributing the user information and the plurality of multimedia information in the recommendation processing request to the M recommendation units, and no intersection exists between the multimedia information distributed by any two recommendation units.

The method for allocating the user information and the plurality of multimedia information in the recommendation processing request to the M recommendation units by the request processing unit for processing is as follows:

in one possible implementation manner, the request processing unit averagely distributes the user information and the plurality of multimedia information in the recommendation processing request to the M recommendation units for processing according to the number of the plurality of multimedia information and the number of the recommendation units.

For example, if the number of multimedia information is 16 and the number of recommendation units is 8, the request processing unit averagely allocates the user information and the plurality of multimedia information in the recommendation processing request to the M recommendation units according to the number of multimedia information and the number of recommendation units: the request processing unit assigns user information and 2 multimedia information to each recommending unit.

In another possible implementation manner, the request processing unit allocates the user information and the plurality of multimedia information according to the load information of each recommending unit, and further allocates the user information and the plurality of multimedia information to the M recommending units for processing according to the load of each recommending unit and a load balancing principle. Correspondingly, the request processing unit acquires the load information of each recommending unit before distributing the user information and the multimedia information, determines a recommending unit with the minimum load from each recommending unit according to the load information of each recommending unit, and distributes the user information and the multimedia information to the recommending unit for processing. Or respectively determining the number of the multimedia information distributed by each recommending unit according to the load information of each recommending unit, so that the recommending unit with a large load distributes a small number of the multimedia information and the recommending unit with a small load distributes a large number of the multimedia information.

The number of the multimedia information can be any number. In this disclosure, taking the number of the multimedia information as 8 and the number of the recommending units as 8 as an example for explanation, correspondingly, each recommending unit may obtain the user characteristics of the user information and the information identifier of part of the multimedia information in the plurality of multimedia information, and an implementation manner of sending the user characteristics and the information identifier of part of the multimedia information in the plurality of multimedia information to the feature service device may be:

the recommending unit 0 acquires the user characteristics according to the user information, acquires the information identifier of the first multimedia information according to the first multimedia information, and sends the user characteristics and the information identifier of the first multimedia information to the characteristic service device. The recommending unit 1 acquires the user characteristics according to the user information, acquires the information identifier of the second multimedia information according to the second multimedia information, and sends the user characteristics and the information identifier of the second multimedia information to the characteristic service device, and so on.

Each recommending unit is used for splicing the characteristic value of the user characteristic returned by the characteristic service device and the characteristic value of the multimedia characteristic of the corresponding multimedia information.

Each recommending unit splices the characteristic value of the user characteristic and the characteristic value of the multimedia characteristic corresponding to the multimedia information, and the way of obtaining the splicing characteristic corresponding to each multimedia information in part of the multimedia information is as follows:

for any recommendation unit and any multimedia information in part of multimedia information corresponding to the recommendation unit, the recommendation unit splices the characteristic value of each user characteristic corresponding to each user characteristic of the user and the characteristic value of each corresponding multimedia characteristic of the multimedia information to obtain the splicing characteristic corresponding to the multimedia information. The present disclosure will be described by taking the recommendation unit 0 as an example.

For example, the number of user features is 3, which are: sex characteristics: "male", occupational characteristics: "athlete", professional characteristics: the feature value of the user characteristic corresponding to the badminton, the feature value of the user characteristic corresponding to the male is ' A ', the feature value of the user characteristic corresponding to the sportsman ' is ' B ', and the feature value of the user characteristic corresponding to the badminton is ' C '. The recommendation unit 0 corresponds to multimedia information, the characteristic value corresponding to the gender information identifier "male" of the multimedia information is "a", the characteristic value corresponding to the professional information identifier "athlete" is "b", and the characteristic value corresponding to the professional information identifier "badminton" is "c". And the recommending unit 0 splices the 'A' and the 'a', the 'B' and the 'B', and the 'C' and the 'C' to obtain the splicing characteristics corresponding to the multimedia information.

In a possible implementation manner, each recommendation unit includes a probability prediction model, and each recommendation unit is configured to input the splicing feature into the probability prediction model and output a matching degree between the user and each piece of multimedia information in the corresponding portion of multimedia information.

Correspondingly, the mode of inputting the splicing characteristics into the probability prediction model by each recommending unit and outputting the matching degree between the user and each multimedia information in the corresponding part of multimedia information is as follows: and each recommending unit inputs the splicing characteristics corresponding to each piece of multimedia information in part of the multimedia information into the probability prediction model to obtain the probability value of the matching between the user information output by the probability prediction model and each piece of multimedia information in the corresponding part of the multimedia information, and the corresponding probability value is used as the matching degree between the user information and each piece of multimedia information in the corresponding part of the multimedia information.

Taking the recommending unit 0 as an example, the recommending unit 0 splices "a" and "a", splices "B" and "B", and splices "C" and "C", so that the obtained splicing characteristics corresponding to the multimedia information are input into the probability prediction model, and the matching degree between the user information output by the probability prediction model and the multimedia information is obtained.

In the embodiment of the disclosure, the recommendation device comprises a plurality of recommendation units, each recommendation unit is only responsible for acquiring the information identifier of part of the multimedia information, so that the plurality of multimedia information are distributed to each recommendation unit, and the plurality of recommendation units perform synchronous processing, thereby improving the service throughput of the recommendation device and reducing the service consumption.

In a possible implementation manner, each recommending unit is further configured to send the matching degree of the user information and each multimedia information in the corresponding part of multimedia information to the request processing unit. The request processing unit is also used for receiving the matching degree of the user information sent by each recommending unit and each piece of multimedia information in part of the multimedia information and outputting the matching degree of the user information and each piece of multimedia information.

In a possible implementation manner, the request processing unit is further configured to sort the plurality of multimedia information according to a matching degree of the user information and each multimedia information, and output the sorted plurality of multimedia information. The manner of ordering the multimedia information by the request processing unit may be: the request processing unit arranges the multimedia information with high matching degree in front and arranges the multimedia information with low matching degree in back according to the matching degree of the user information and each multimedia information. For example, if the matching degree is a probability value of matching the user information with the multimedia information, the request processing unit ranks the multimedia information with a high corresponding probability value in front of the request processing unit and ranks the multimedia information with a low corresponding probability value in back of the request processing unit.

In the embodiment of the disclosure, the request processing unit sequences the plurality of multimedia information by receiving the matching degree between the user information sent by each recommending unit and each multimedia information in the corresponding part of multimedia information according to the matching degree between the user information and each multimedia information, and outputs the sequenced plurality of multimedia information, thereby facilitating the multimedia server to recommend the multimedia information according to the output of the multimedia recommending system.

It should be noted that, in the present disclosure, the recommendation device processes the total information in the recommendation processing request, that is, all the user information and the multimedia information, and each recommendation unit processes only a part of the total information, and there is no intersection between the information processed by each recommendation unit, so that the information to be processed in the recommendation processing request is distributed to a plurality of recommendation units, and the plurality of recommendation units process synchronously, thereby greatly increasing the speed of information processing, and thus increasing the multimedia recommendation efficiency of the multimedia recommendation system.

Referring to fig. 5, the feature service device includes N feature service units, where N is an integer greater than 1; each feature service unit is connected with each recommending unit.

and the characteristic service unit is also used for reading the multimedia characteristics corresponding to the information identification of the received multimedia information from part of the stored multimedia characteristics and returning the read multimedia characteristics to the corresponding recommendation unit.

The number of the feature service units and the number of the recommending units can be the same or different. In addition, the number of the recommending unit and the feature serving unit may be configured as needed, and in the embodiment of the present disclosure, the number of the recommending unit and the feature serving unit is the same and is 8, that is, M and N are both equal to 8. For example, with continued reference to FIG. 5, the number of recommendation units is 8, recommendation unit 0, recommendation unit 1,. cndot. cndot.7, respectively. The number of feature service units is 8, respectively feature service unit 0, feature service unit 1,. cndot. cndot.7. The feature service unit is a Parameter shard obtained by fragmenting the feature service device.

Wherein, each recommendation unit is connected with each feature service unit, the recommendation unit 0 is respectively connected with the feature service unit 0, the feature service unit 1, the feature service unit 7, the recommendation unit 1 is respectively connected with the feature service unit 0, the feature service unit 1, the feature service unit 7, and so on.

In the embodiment of the disclosure, each feature service unit stores part of features in the full amount of multimedia features, and there is no repeated feature between the features stored in each feature service unit and the features stored in other feature service units, so that the online feature service device can store more features, thereby improving the accuracy of multimedia recommendation and ensuring that the online profit is the same as the profit of offline training. For example, the number of offline features trained by the offline feature training device is 340 hundred million, in the prior art, only 60 hundred million features can be stored online, and in the present disclosure, when the number of feature service units is 8, each feature service unit only needs to store 42.5 million features, so that the number of online features is the same as that of offline features, and the online profit and the offline profit are ensured to be consistent.

In a possible implementation manner, when the feature service device includes a plurality of feature service units, an embodiment in which each recommending unit sends the user feature and the information identifier of the corresponding part of the multimedia information in the plurality of multimedia information to the feature service device may be: and each recommending unit sends the user characteristics and the information identification of the part of the multimedia information corresponding to the recommending unit to each characteristic service unit.

For example, in the embodiment of the present disclosure, it is explained that the number of recommendation units is 8, and the number of feature service units is 8, and the recommendation unit 0 sends the acquired user feature and the information identifier of the corresponding part of the multimedia information to the feature service unit 0, the feature service unit 1, · · · and the feature service unit 7, respectively. The recommendation unit 1 sends the obtained user characteristics and the information identifications of the corresponding part of the multimedia information to the characteristic service unit 0, the characteristic service unit 1, the characteristic service unit 7, and so on.

And the characteristic service unit is used for receiving the information identifier of each piece of multimedia information in the corresponding part of multimedia information sent by each recommendation unit, reading the multimedia characteristics corresponding to the information identifier of the received multimedia information from the part of multimedia characteristics stored in the characteristic service unit according to the received information identifier of each piece of multimedia information, and returning the read multimedia characteristics to the corresponding recommendation unit.

In one possible implementation, each feature service unit stores a correspondence between an information identifier and a feature value of each multimedia feature in the partial multimedia feature. Correspondingly, the manner of reading the multimedia features corresponding to the information identifier of the received multimedia information from the part of the stored multimedia features by the feature service unit according to the information identifier of each received multimedia information is as follows: and the characteristic service unit reads the characteristic value of the multimedia characteristic of each piece of multimedia information from the corresponding relation between the information identification and the characteristic value according to the received information identification of each piece of multimedia information.

The characteristic service unit is also used for storing the corresponding relation between part of the user characteristics and the characteristic values. The characteristic service unit is also used for receiving the corresponding user characteristics sent by each recommending unit and reading the characteristic values corresponding to the user characteristics from the corresponding relation between the user characteristics and the characteristic values according to the received user characteristics.

Correspondingly, for the information identifier corresponding to any multimedia information, if the feature service unit stores the feature value of the multimedia feature corresponding to the information identifier, the feature value of the multimedia feature corresponding to the information identifier is returned to the recommending unit which sends the information identifier; and if the characteristic service unit stores the characteristic value corresponding to the user characteristic, returning the characteristic value of the user characteristic to the recommending unit which sends the user characteristic.

It should be noted that the correspondence between the information identifier and the feature value may be a key-value pair. The key is the information identification of the multimedia characteristic corresponding to the key-value pair, and the value is the characteristic value of the multimedia characteristic corresponding to the key-value pair.

In a possible implementation manner, the offline feature training device is further configured to determine, when a multimedia feature is added, a feature service unit in which the multimedia feature is stored, and store the multimedia feature in a corresponding feature service unit. The implementation of this process may be:

when the multimedia features are added, the offline feature training device performs hash operation on the information identification of the multimedia features to obtain hash values of the multimedia features, maps the number N of the feature service units according to the hash values to obtain feature service units with stored multimedia features, and sends the multimedia features to corresponding feature service units.

In one possible implementation manner, the manner in which the feature service unit stores the corresponding multimedia feature may be: the feature service unit encodes the multimedia feature into a predefined format, thereby converting the multimedia feature into a key-value pair (key-value pair), and stores the key-value pair.

In the embodiment of the disclosure, when a multimedia feature is added, the offline feature training device determines the feature service unit in which the multimedia feature is stored, only the feature service unit stores the multimedia feature, and each feature training module does not store repeated features, so that the feature quantity of the online feature service device is enlarged, and the online business index is improved. And the off-line characteristic training device only needs to transmit the characteristics to the characteristic service unit and does not need to transmit the characteristics to the recommendation unit, so that the pressure of data transmission is reduced by 83.5 percent.

Fig. 6 is a flowchart of a multimedia recommendation method provided in an embodiment of the present disclosure. Referring to fig. 6, the multimedia recommendation method may be applied in a recommendation device, and the embodiment includes:

in step 601, a recommendation processing request carrying user information and a plurality of multimedia information is received, a user characteristic corresponding to the user information is obtained, and an information identifier of the multimedia information is sent to the characteristic service device.

In step 602, the user characteristics and the multimedia characteristics read by the characteristic service device are spliced to obtain splicing characteristics, and the matching degree between the user information and each multimedia information is obtained according to the splicing characteristics.

In the embodiment of the disclosure, the recommendation device sends the information identifier of the multimedia information to the feature service device, and obtains the matching degree between the user information and the multimedia information based on the multimedia features read by the feature service device, so that the recommendation device does not need to store the multimedia features and read the features, the capacity of the recommendation device is greatly saved, and the efficiency of obtaining the matching degree between the user information and each multimedia information by the recommendation device can be improved.

the method for sending the information identification of the multimedia information to the feature service device comprises the following steps:

distributing a plurality of multimedia information to M recommending units, wherein no intersection exists between the multimedia information distributed by any two recommending units, and respectively sending information identifiers of the multimedia information to a feature service device through the M recommending units;

splicing the user characteristics and the multimedia characteristics read by the characteristic service device to obtain splicing characteristics, and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics, wherein the method comprises the following steps:

splicing the user characteristics and the multimedia characteristics of the corresponding multimedia information returned by the characteristic service device through M recommending units to obtain splicing characteristics; and acquiring the matching degree of the user information and the corresponding multimedia information according to the splicing characteristics.

In another possible implementation manner, obtaining the matching degree between the user information and each multimedia information according to the splicing characteristics includes:

In another possible implementation, distributing the plurality of multimedia information to M recommendation units includes:

averagely distributing a plurality of multimedia information to M recommending units; or distributing the plurality of multimedia information to the M recommending units according to the load of each recommending unit and the load balancing principle.

Fig. 7 is a flowchart of a multimedia recommendation method provided in an embodiment of the present disclosure. Referring to fig. 7, the multimedia recommendation method may be applied to a feature service device, and the method includes:

in step 701, an information identifier of each multimedia information sent by the recommending apparatus is received.

In step 702, according to the information identifier of each multimedia message, the multimedia feature corresponding to the information identifier of each multimedia message is read.

In step 703, the read multimedia features are returned to the recommendation device.

In the embodiment of the disclosure, the feature service device reads the corresponding multimedia features according to the information identifier of the multimedia information sent by the recommendation device, and returns the read multimedia features to the recommendation device, so that the recommendation device does not need to store and read the features, the capacity of the recommendation device is saved, and the information processing efficiency of the recommendation device is improved. On the other hand, the feature service device reads the features according to the information identifier and then returns the features to the sender of the information identifier, so that independent feature reading is realized, and therefore, feature reading services can be provided for different recommendation devices.

reading the multimedia characteristics corresponding to the information identification of each multimedia information according to the information identification of each multimedia information, comprising:

returning the read multimedia features to the recommendation device, comprising:

and respectively returning the read multimedia characteristics to the recommendation device through the N characteristic service units.

Fig. 8 is a flowchart of a multimedia recommendation method provided in an embodiment of the present disclosure. Referring to fig. 8, the multimedia recommendation method may be applied to an offline feature training apparatus, and the method includes:

in step 801, training multimedia features, synchronizing the multimedia features to a feature service device; the feature service device comprises N feature service units, wherein N is an integer greater than 1; each feature service unit stores part of the multimedia features, and no intersection exists between the part of the multimedia features stored by any two feature service units.

In step 802, when a multimedia feature is added, a feature service unit in which the multimedia feature is stored is determined.

In step 803, the multimedia feature is stored to the determined feature service unit.

In the disclosed embodiment, the offline feature training device synchronizes the multimedia features to the online feature service device by training the multimedia features, thereby continuously increasing the storage capacity of the online features. On the other hand, the off-line feature training device determines the feature service units with the stored multimedia features and stores the multimedia features into the determined feature service units when the multimedia features are added, so that repeated features are not stored among the feature service units of the on-line feature service device, the storage amount of the on-line features is greatly improved, and the accuracy of information recommendation of a subsequent multimedia recommendation system is guaranteed when the multimedia recommendation is performed based on the on-line multimedia features.

In one possible implementation, when a multimedia feature is added, a feature service unit that determines that the multimedia feature is stored includes:

when the multimedia features are added, performing hash operation on the multimedia features to obtain hash values of the multimedia features; and mapping the N according to the hash value to obtain a feature service unit with stored multimedia features.

Fig. 9 is a block diagram of a recommendation device provided in an embodiment of the present disclosure. Referring to fig. 9, the apparatus includes:

a recommendation processing request receiving module 901 configured to execute a recommendation processing request carrying user information and a plurality of multimedia information.

The user characteristic obtaining module 902 is configured to perform obtaining of a user characteristic corresponding to the user information.

An information identifier sending module 903 configured to execute sending the information identifier of the multimedia information to the feature service device.

And a matching degree obtaining module 904, configured to perform splicing of the user characteristics and the multimedia characteristics read by the characteristic service device to obtain splicing characteristics, and obtain matching degrees between the user information and each piece of multimedia information according to the splicing characteristics.

the information identifier sending module is also configured to allocate the plurality of multimedia information to the M recommending units, no intersection exists between the multimedia information allocated by any two recommending units, and the M recommending units are used for sending the information identifiers of the multimedia information to the feature service device respectively;

a matching degree obtaining module 904 configured to perform splicing of the user characteristics and the multimedia characteristics of the corresponding multimedia information returned by the characteristic service device by the M recommending units, respectively, to obtain spliced characteristics; and acquiring the matching degree of the user information and each multimedia information according to the splicing characteristics.

In another possible implementation manner, the matching degree obtaining module 904 is configured to input the splicing features into the probability prediction model to obtain the matching degree between the user information and each multimedia information.

In another possible implementation manner, the information identifier sending module is configured to perform average distribution of the plurality of multimedia information to M recommendation units; or distributing the plurality of multimedia information to the M recommending units according to the load of each recommending unit and the load balancing principle.

In another possible implementation manner, the apparatus further includes:

Fig. 10 is a block diagram of a feature service apparatus provided in an embodiment of the present disclosure. Referring to fig. 10, the apparatus includes:

an information identifier receiving module 1001 configured to perform receiving information identifiers of each multimedia information transmitted by the recommending apparatus.

The multimedia feature reading module 1002 is configured to perform reading of a multimedia feature corresponding to an information identifier of each multimedia message according to the information identifier of each multimedia message.

A multimedia feature transmitting module 1003 configured to perform returning the read multimedia features to the recommending device.

a multimedia feature reading module 1002 configured to perform reading, by the N feature service units, a multimedia feature corresponding to an information identifier of the multimedia information from a part of the multimedia features stored therein, respectively;

a multimedia feature transmitting module 1003 configured to execute returning the read multimedia features to the recommending device through the N feature service units, respectively.

Fig. 11 is a block diagram of an offline feature training apparatus provided in an embodiment of the present disclosure. Referring to fig. 11, the apparatus includes:

a feature training module 1101 configured to perform training of multimedia features, synchronizing the multimedia features into a feature service device; the feature service device comprises N feature service units, wherein N is an integer greater than 1; each feature service unit stores part of the multimedia features, and no intersection exists between the part of the multimedia features stored by any two feature service units.

A feature service unit determination module 1102 configured to perform determining a feature service unit in which the multimedia feature is stored when the multimedia feature is added.

A feature storage module 1103 configured to perform storing the multimedia features to the determined feature service unit.

In a possible implementation manner, the feature service unit determining module 1102 is configured to perform a hash operation on the multimedia features when the multimedia features are added, so as to obtain a hash value of the multimedia features; and mapping the N according to the hash value to obtain a feature service unit with stored multimedia features.

All the above optional technical solutions may be combined arbitrarily to form the optional embodiments of the present disclosure, and are not described herein again.

It should be noted that: in the above embodiment, when performing multimedia recommendation, the recommendation apparatus, the feature service apparatus, and the offline feature training apparatus are exemplified by only dividing the functional modules, and in practical applications, the functions may be distributed by different functional modules according to needs, that is, the internal structure of the device may be divided into different functional modules to complete all or part of the functions described above.

Fig. 12 is a schematic structural diagram of a multimedia recommendation apparatus 1200, where the multimedia recommendation apparatus 1200 may generate a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 1201 and one or more memories 1202, where the memory 1202 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 1201 to implement the methods provided by the above method embodiments. Certainly, the multimedia recommendation device 1200 may further include components such as a wired or wireless network interface, a keyboard, and an input/output interface, so as to perform input and output, and the multimedia recommendation device 1200 may further include other components for implementing device functions, which are not described herein again.

In an exemplary embodiment, a computer-readable storage medium, such as a memory, including instructions executable by a processor in a terminal to perform the multimedia recommendation method in the above embodiments is also provided. For example, the computer readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

The above description is only exemplary of the present disclosure and is not intended to limit the present disclosure, so that any modification, equivalent replacement, or improvement made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.

Claims

1. A multimedia recommendation system, comprising: a recommendation device and a feature service device;

2. The system of claim 1, wherein the recommendation device comprises M recommendation units, M being an integer greater than 1;

3. The system of claim 2, wherein the feature service device comprises N feature service units, N being an integer greater than 1;

4. A multimedia recommendation method is applied to a recommendation device, and comprises the following steps:

5. A multimedia recommendation method, wherein the method is applied in a feature service device, and the method comprises:

and returning the read multimedia features to the recommending device.

6. A multimedia recommendation method is applied to an off-line feature training device, and comprises the following steps:

7. A recommendation device, characterized in that the device comprises:

8. An aspect service apparatus, the apparatus comprising:

9. An offline feature training device, the device comprising:

10. A multimedia recommendation apparatus, characterized in that the apparatus comprises: one or more processors;

volatile or non-volatile memory for storing the one or more processor-executable instructions;

wherein the one or more processors are configured to perform the multimedia recommendation method of any of claims 4-6.