CN111488526B - Recommendation method and device - Google Patents
Recommendation method and device Download PDFInfo
- Publication number
- CN111488526B CN111488526B CN202010289694.0A CN202010289694A CN111488526B CN 111488526 B CN111488526 B CN 111488526B CN 202010289694 A CN202010289694 A CN 202010289694A CN 111488526 B CN111488526 B CN 111488526B
- Authority
- CN
- China
- Prior art keywords
- information
- user
- sub
- feature
- vectors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 239000013598 vector Substances 0.000 claims abstract description 365
- 230000006399 behavior Effects 0.000 claims description 106
- 108091026890 Coding region Proteins 0.000 claims description 14
- 230000003068 static effect Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 9
- 238000012549 training Methods 0.000 description 9
- 230000004927 fusion Effects 0.000 description 8
- 238000013507 mapping Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3438—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment monitoring of user actions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- Computer Hardware Design (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a recommendation method and a recommendation device, wherein the recommendation method comprises the following steps: acquiring first type characteristic information of a user; generating a first feature vector corresponding to the first type of feature information, wherein the user portrait of the user is represented by the first feature vector; calculating the similarity between the first feature vector and a third feature vector corresponding to the candidate recommended object; and recommending the target recommended objects with the similarity meeting the preset condition in the candidate recommended objects to the user. According to the embodiment of the invention, the user portraits represented by the continuous feature vectors are not fractured due to the correlation among the features of different dimensions of the user, so that when the object recommendation is performed based on the user feature vectors, the recommended object and the comprehensive features of the dimensions of the user can be ensured to be more matched, and further a better recommendation effect can be ensured.
Description
Technical Field
The present invention relates to the field of information processing technologies, and in particular, to a recommendation method and apparatus.
Background
User portraits can be simply understood as labels of massive data, and the principle is that users are divided into different types according to the differences of attributes, behaviors, views and the like of the users, then typical features are extracted from each type, and descriptions of names, photos, some demographic factors, scenes and the like are given to form a personal prototype (Personas). The constructed user image can be used in various fields such as accurate marketing, user statistics, data mining, private customization and the like.
The current user portrait usually adopts a mode of combining dynamic information data (such as user operation behavior data) with static information data (such as user personal information), and when in use, the labels of the user portrait are regarded as discrete attribute values, and targeted user recommendation and analysis are performed by a similarity measurement mode.
In practical situations, the labels in the user portrait generally have correlation, and the discrete code cannot express the correlation between the labels, for example, the gender label may have strong correlation with the music label, so that the discrete label expression method can miss more information.
Therefore, the conventional user portrait construction method breaks the correlation among user labels, so that the recommendation effect is poor when the related objects are recommended based on the user portrait.
Disclosure of Invention
The embodiment of the invention provides a recommendation method and device, which are used for solving the problem of poor recommendation effect when a related object is recommended based on a user portrait.
In order to solve the technical problems, the invention is realized as follows:
in a first aspect, an embodiment of the present invention provides a recommendation method, including:
acquiring first type characteristic information of a user, wherein the first type characteristic information comprises N pieces of characteristic information, and N is an integer greater than 1;
Generating a first feature vector corresponding to the first type of feature information, wherein the first feature vector is an N-dimensional space vector, different dimensions in the N-dimensional space correspond to different feature information in the N feature information, and a user portrait of the user is represented by the first feature vector;
calculating the similarity between the first feature vector and a third feature vector corresponding to the candidate recommended object;
and recommending the target recommended objects with the similarity meeting the preset condition in the candidate recommended objects to the user.
Optionally, the first type of feature information is static feature information of the user; the method further comprises the steps of:
acquiring second type characteristic information of the user, wherein the second type characteristic information comprises M pieces of characteristic information, M is an integer greater than 1, and the second type characteristic information is dynamic behavior characteristic information of the user;
generating a second feature vector corresponding to the second type of feature information;
splicing the second characteristic vector and the first characteristic vector to obtain a target characteristic vector, wherein the user portrait of the user is represented by the target characteristic vector;
the calculating the similarity between the first feature vector and the third feature vector corresponding to the candidate recommended object includes:
And calculating the similarity between the target feature vector and a third feature vector corresponding to the candidate recommended object.
Optionally, the obtaining the second type of feature information of the user includes:
obtaining K pieces of operation object information of the user, wherein the K pieces of operation object information are information of K operation objects aimed at by the operation behaviors of the user, and K is an integer greater than or equal to 1;
respectively acquiring operation behavior information of the user on each operation object in the K operation objects to acquire M operation behavior information;
the generating the second feature vector corresponding to the second type of feature information includes:
respectively generating first sub-vectors corresponding to each piece of operation object information in the K pieces of operation object information to obtain K pieces of first sub-vectors;
generating first sub-vectors corresponding to each piece of operation behavior information in the M pieces of operation behavior information respectively to obtain M first sub-vectors;
respectively splicing first sub-vectors corresponding to the operation behavior information of the same operation object in the M first sub-vectors to obtain K second sub-vectors;
respectively splicing the first sub-vectors and the second sub-vectors corresponding to the same operation object in the K first sub-vectors and the K second sub-vectors to obtain K third sub-vectors;
And splicing each third component in the K third component vectors to obtain the second feature vector corresponding to the second type feature information.
Optionally, the generating the first feature vector corresponding to the first type of feature information includes:
generating second sub-vectors corresponding to each piece of characteristic information in the N pieces of characteristic information respectively, wherein each second sub-vector corresponds to different dimensions respectively;
and splicing each second sub-vector to obtain the first feature vector corresponding to the first type of feature information.
Optionally, the generating a second sub-vector corresponding to each of the N pieces of feature information includes:
respectively encoding each piece of characteristic information in the N pieces of characteristic information to obtain a coding sequence corresponding to each piece of characteristic information;
and generating second sub-vectors of corresponding dimensions from the coding sequences corresponding to each piece of characteristic information by adopting a word embedding mode.
In a second aspect, an embodiment of the present invention provides a recommendation apparatus, including:
the device comprises a first acquisition module, a second acquisition module and a first processing module, wherein the first acquisition module is used for acquiring first type characteristic information of a user, the first type characteristic information comprises N pieces of characteristic information, and N is an integer larger than 1;
The first generation module is used for generating a first feature vector corresponding to the first type of feature information, wherein the first feature vector is an N-dimensional space vector, different dimensions in the N-dimensional space correspond to different feature information in the N feature information, and a user portrait of the user is represented through the first feature vector;
the computing module is used for computing the similarity between the first feature vector and a third feature vector corresponding to the candidate recommended object;
and the recommending module is used for recommending target recommended objects with similarity meeting preset conditions in the candidate recommended objects to the user.
Optionally, the first type of feature information is static feature information of the user; the recommendation device further includes:
the second acquisition module is used for acquiring second type characteristic information of the user, wherein the second type characteristic information comprises M pieces of characteristic information, M is an integer greater than 1, and the second type characteristic information is dynamic behavior characteristic information of the user;
the second generation module is used for generating a second feature vector corresponding to the second type of feature information;
the splicing module is used for splicing the second characteristic vector and the first characteristic vector to obtain a target characteristic vector, wherein the user portrait of the user is represented by the target characteristic vector;
The calculation module is used for calculating the similarity between the target feature vector and a third feature vector corresponding to the candidate recommended object.
Optionally, the second obtaining module includes:
a first obtaining unit, configured to obtain K pieces of operation object information of the user, where the K pieces of operation object information are information of K operation objects aimed at by an operation behavior of the user, and K is an integer greater than or equal to 1;
the second acquisition unit is used for respectively acquiring the operation behavior information of the user on each operation object in the K operation objects to obtain M operation behavior information;
the second generation module includes:
the first generation unit is used for respectively generating first sub-vectors corresponding to each piece of operation object information in the K pieces of operation object information to obtain K pieces of first sub-vectors;
the second generating unit is used for respectively generating first sub-vectors corresponding to each piece of operation behavior information in the M pieces of operation behavior information to obtain M pieces of first sub-vectors;
the first splicing unit is used for respectively splicing the first sub-vectors corresponding to the operation behavior information of the same operation object in the M first sub-vectors to obtain K second sub-vectors;
The second splicing unit is used for respectively splicing the first sub-vectors and the second sub-vectors corresponding to the same operation object in the K first sub-vectors and the K second sub-vectors to obtain K third sub-vectors;
and the third splicing unit is used for splicing each third component in the K third component vectors to obtain the second characteristic vector corresponding to the second type characteristic information.
Optionally, the first generating module includes:
the third generating unit is used for respectively generating second sub-vectors corresponding to each piece of characteristic information in the N pieces of characteristic information, and each second sub-vector corresponds to different dimensions;
and the fourth splicing unit is used for splicing each second sub-vector to obtain the first feature vector corresponding to the first type of feature information.
Optionally, the third generating unit includes:
the coding subunit is used for respectively coding each piece of characteristic information in the N pieces of characteristic information to obtain a coding sequence corresponding to each piece of characteristic information;
and the generation subunit is used for respectively generating second sub-vectors of corresponding dimensions from the coding sequences corresponding to the characteristic information by adopting a word embedding mode.
In a third aspect, an embodiment of the present invention provides a recommendation apparatus, including a processor, a memory, and a computer program stored on the memory and executable on the processor, where the computer program when executed by the processor implements the steps in the recommendation method described above.
In a fourth aspect, embodiments of the present invention provide a computer readable storage medium having a computer program stored thereon, which when executed by a processor, implements the steps of the above-mentioned recommendation method.
In the embodiment of the invention, the target recommended object with the similarity meeting the preset condition is recommended to the user by mapping the characteristic information of different dimensions of the user into the continuous characteristic vector and by comparing the similarity between the characteristic vector of the user and the characteristic vector of the candidate recommended object. Therefore, the user portraits represented by the continuous feature vectors do not fracture the correlation among the features of different dimensions of the user, so that when the object recommendation is performed based on the user feature vectors, the recommended object and the comprehensive features of the dimensions of the user can be ensured to be more matched, and further, a good recommendation effect can be ensured.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort to a person of ordinary skill in the art.
FIG. 1 is a flowchart of a recommendation method provided in an embodiment of the present invention;
FIG. 2 is an exemplary flow chart of a recommendation method provided by an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a recommendation device according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, fig. 1 is a flowchart of a recommendation method provided in an embodiment of the present invention, which is applied to a recommendation device, as shown in fig. 1, and the method includes the following steps:
Step 101, obtaining first type feature information of a user, wherein the first type feature information comprises N pieces of feature information, and N is an integer greater than 1.
The first type of feature information may include feature information of the user in a certain aspect, for example, may include personal information of the user, such as age, gender, region, occupation, etc., or may include operation behavior information of the user, such as browsing a web page, playing a song, publishing an article, praying, etc., and in particular, may acquire operation behavior information of the user in a certain aspect, for example, may acquire operation behavior information of the user in listening to music, such as playing a song, collecting a song, sharing a song, praying a song, etc. Of course, the first type of characteristic information may also include both personal information of the user and operational behavior information of the user.
The first type of feature information of the user may be obtained through big data statistics, for example, for personal information of the user, the first type of feature information may be obtained through obtaining personal information provided when the user registers, and for operational behavior information of the user, the first type of feature information may be obtained through recording operational behavior information of the user on a web page or a specific application program.
It should be noted that, in the embodiment of the present invention, the first type of feature information may include N feature information, where each feature information may represent a feature of one dimension, that is, may include feature information of the user in multiple dimensions, but is not limited to a certain feature information, and in order to ensure that a user portrait with a comprehensive feature coverage is constructed as much as possible, feature information of each different dimension of the user may be acquired as much as possible.
Step 102, generating a first feature vector corresponding to the first type of feature information, wherein the first feature vector is an N-dimensional space vector, different dimensions in the N-dimensional space correspond to different feature information in the N feature information, and a user portrait of the user is represented by the first feature vector.
The generating the first feature vector corresponding to the first type of feature information may refer to mapping the first type of feature information into feature vectors of a multidimensional space, so as to represent feature information of N different dimensions of the user through a continuous feature vector, and thus, a value of the generated first feature vector mapped on a certain dimension represents a value corresponding to the feature information of the user on the dimension.
In the embodiment of the invention, the first feature vector can be used for representing the user portrait of the user, so that the feature information of the user in different dimensions can be fused through vectors instead of discrete feature values, and the feature information with relevance can be prevented from being split.
The generating the first feature vector corresponding to the first type of feature information may include:
generating second sub-vectors corresponding to each piece of characteristic information in the N pieces of characteristic information respectively, wherein each second sub-vector corresponds to different dimensions respectively;
and splicing each second sub-vector to obtain the first feature vector corresponding to the first type of feature information.
Specifically, the generating the first feature vector corresponding to the first type of feature information may be generating respective corresponding second sub-vectors for each feature information in the N feature information, to obtain N second sub-vectors, and then splicing each second sub-vector in the N second sub-vectors to obtain the first feature vector.
The generating of the second sub-vector corresponding to each piece of feature information may be allocating a dimension to each piece of feature information, and then determining a value of each piece of feature information in each dimension, so as to obtain a second sub-vector corresponding to each piece of feature information, where each obtained second sub-vector corresponds to a different dimension, and specifically, a vector corresponding to each piece of feature information may be generated by embedding in an embedding manner.
More specifically, the vector generation model may be used to perform the vector generation operation on each feature information, that is, the corresponding pre-trained vector generation model may be obtained for each dimension corresponding to each feature information, and then the corresponding second sub-vector may be generated and obtained only by inputting each feature information into the corresponding vector generation model. The vector generation model can be obtained by acquiring a feature information training set of corresponding dimensions, and training an initial vector generation model by taking the vector of the corresponding dimensions as a target.
The above-mentioned process of stitching each second sub-vector, that is, performing feature fusion on each feature information, may specifically be that adding is performed on each second sub-vector to obtain an N-dimensional vector, that is, the first feature vector.
In this way, the first feature vector corresponding to the first type of feature information can be obtained by respectively generating the second sub-vector corresponding to each feature information in the N feature information and then splicing each second sub-vector, so that feature fusion of the first type of feature information is realized, and the correlation among the feature information is reserved.
Further, the generating the second sub-vector corresponding to each of the N pieces of feature information includes:
respectively encoding each piece of characteristic information in the N pieces of characteristic information to obtain a coding sequence corresponding to each piece of characteristic information;
and generating second sub-vectors of corresponding dimensions from the coding sequences corresponding to each piece of characteristic information by adopting a word embedding mode.
In this embodiment, the encoding of each of the N pieces of feature information may be performed by encoding each of the N pieces of feature information by using one-hot vector encoding (one-hot), to obtain a coding sequence corresponding to each piece of feature information. For example, for sex characteristic information of a user, "01" may be used to indicate a sex male and "10" may be used to indicate a sex female, and if the sex of the user is female, a corresponding code sequence "10" may be obtained by performing one-hot encoding on the sex characteristic of the user.
For another example, for the operational behavior characteristic information of the user in listening to a song, such as playing, collecting, praying, sharing and dislike five behavior characteristics, the playing behavior may be represented by "00001", the collecting behavior may be represented by "00010", the praying behavior may be represented by "00100", the sharing behavior may be represented by "01000", and the dislike behavior may be represented by "10000", and if the user collects a certain song, the one-hot encoding may be performed on the behavior characteristic of the user in collecting the song, so as to obtain the corresponding encoding sequence "00010".
Then, a word embedding (word embedding) manner may be adopted to generate the second sub-vectors of the corresponding dimensions from the coding sequences corresponding to each piece of feature information, specifically, the coding sequences corresponding to each piece of feature information may be multiplied by corresponding lookup tables respectively to obtain the corresponding second sub-vectors, where the lookup tables may be a matrix formed by multidimensional vectors. For example, the lookup table corresponding to the operation behavior of listening to songs is a 5×1 matrix, that is, the matrix includes 5 rows and 1 column, and the values of each row are different, and for the coding sequence "00010" corresponding to the behavior of collecting songs, the row vector [ 00010 ] may be used to multiply the matrix to obtain the row vector corresponding to the fourth row in the matrix, that is, the second sub-vector corresponding to the behavior of collecting songs. The lookup table may be a pre-trained lookup table capable of reflecting the feature information of the user more accurately, that is, the determination of each value is trained based on the feature data of the user.
For example, for a coding sequence "01" corresponding to a gender feature of a user, a sub-vector of the gender dimension may be generated, such as a sub-vector of length 1 and positive in that dimension, while for a coding sequence "01", a sub-vector of length 1 and negative in that dimension may be generated; for the coding sequence "00001" corresponding to the operation behavior feature of a song played by a user, a subvector with a feature dimension corresponding to the song can be generated, for example, a subvector with a positive direction and a length of 0.5 in the dimension is generated, for the coding sequence "00010", a subvector with a positive direction and a length of 1 in the dimension can be generated, for the coding sequence "00100", a subvector with a positive direction and a length of 1.5 in the dimension can be generated, for the coding sequence "01000", a subvector with a positive direction and a length of 2 in the dimension can be generated, and for the coding sequence "10000", a subvector with a negative direction and a length of 1 in the dimension can be generated, so that the polarity progression of the preference of the song by the user can be reflected through the vector corresponding to each different operation behavior of the user in listening to the song can be reflected, namely, the greater the vector value of the dimension indicates the greater preference degree of the song by the user.
In this way, each feature information in the N feature information is encoded to obtain a corresponding encoded sequence, and then each encoded sequence is generated into a second sub-vector with a corresponding dimension by adopting a word embedding mode, so that the feature mapping of each user with different dimensions into a continuous vector can be realized to realize feature fusion.
And step 103, calculating the similarity between the first feature vector and a third feature vector corresponding to the candidate recommended object.
The candidate recommended objects may be various objects to be recommended to the user, such as some physical commodities, virtual services, etc., for example, sound boxes, mobile phones, music, application programs, financial products, etc.
The third feature vector corresponding to the candidate recommended object may be a feature vector generated according to the attribute feature of the candidate recommended object, and may specifically be generated by using a text-vector doc2vec algorithm. For example, if the candidate recommended object is a song, a feature vector corresponding to the song may be generated according to attribute features of a singer, a song wind, a year, a language, etc. of the song, where each dimension of the feature vector represents an attribute feature of the song, and a specific generation manner may be similar to a vector generation manner of the first type of feature information, which is not described herein.
The calculating the similarity between the first feature vector and the third feature vector corresponding to the candidate recommended object may be calculating the cosine similarity between the first feature vector and the third feature vector, that is, calculating the cosine value of the included angle between the two feature vectors to obtain the similarity, where the cosine value ranges between [ -1,1] and the closer the value is to 1, the closer the directions of the two vectors are represented, that is, the greater the similarity is, the closer to 100%; the more toward-1, the more opposite their direction, i.e., the less similarity, the more toward 0%; a value close to 0 indicates that the two vectors are nearly orthogonal, i.e. the similarity is close to 50%.
More specifically, the similarity between vectors can be quickly and accurately calculated by training a similarity calculation model, and the training process of the similarity calculation model can be as follows: and obtaining a large number of feature vectors corresponding to the user feature information and feature vectors corresponding to the pre-recommended objects as an input data training set, wherein the feature vectors corresponding to the user feature information and the feature vectors corresponding to each pre-recommended object are calibrated with similarity, training a cosine similarity calculation model by using the input data training set, comparing the similarity output by the model with the calibrated similarity, calculating the mean square error of the two, adjusting and correcting parameters of the cosine similarity calculation model under the condition of large error, and repeating the training process until the error is in a preset range and tends to be stable, thereby obtaining the similarity calculation model after training. In this way, when the similarity between the first feature vector and the third feature vector corresponding to the candidate recommended object is calculated, the first feature vector and the third feature vector can be input into the similarity calculation model, so that the similarity output by the similarity calculation model can be obtained quickly.
It should be noted that the first feature vector and the third feature vector may have a relatively consistent spatial dimension, that is, the types and the number of feature information are relatively consistent, so as to ensure that an object conforming to the preference or the use habit of the user can be recommended to the user based on the similarity of the first feature vector and the third feature vector. For example, the first feature vector of the user may be generated based on information such as age, gender, song listening habit or preference of the user, and the third feature vector may be generated based on feature information such as singer, singer gender, age, and song wind of candidate recommended songs, so that based on the feature vectors of the two, whether the song accords with the song listening preference of the user can be reflected more truly.
Optionally, the first type of feature information is static feature information of the user; the method further comprises the steps of:
acquiring second type characteristic information of the user, wherein the second type characteristic information comprises M pieces of characteristic information, M is an integer greater than 1, and the second type characteristic information is dynamic behavior characteristic information of the user;
generating a second feature vector corresponding to the second type of feature information;
splicing the second characteristic vector and the first characteristic vector to obtain a target characteristic vector, wherein the user portrait of the user is represented by the target characteristic vector;
The calculating the similarity between the first feature vector and the third feature vector corresponding to the candidate recommended object includes:
and calculating the similarity between the target feature vector and a third feature vector corresponding to the candidate recommended object.
In the embodiment, in order to more refine and comprehensively construct the user portrait, the feature information of different types of users can be respectively obtained, corresponding feature vectors are respectively generated for the feature information of different types, and then the feature vectors are spliced to ensure that the user portrait with stronger descriptive capacity, namely, the user feature attribute can be more comprehensively reflected.
Specifically, the first type of feature information may be static feature information of the user, such as personal information of the user, including age, gender, region, occupation, etc., and the second type of feature information may be dynamic behavior feature information of the user, such as operational behavior information of the user, including browsing web pages, playing songs, publishing articles, praise, etc., and in particular, may be obtaining operational behavior information of the user in a certain aspect, for example, may obtain operational behavior information of the user in listening to music, such as playing songs, sharing songs, praise songs, etc.
Similarly, the second type of feature information of the user can be obtained by recording operation behavior information of the user on a webpage or a specific application program.
It should be noted that, in the embodiment of the present invention, the second type of feature information may include M feature information, where each different feature information may represent a feature of a different dimension, that is, may include operation behavior feature information of the user in one or more dimensions, but is not limited to a certain feature information, and in order to ensure that a user portrait with a relatively comprehensive feature coverage is constructed as much as possible, operation behavior feature information of each different dimension of the user may be acquired as much as possible.
The generating the second feature vector corresponding to the second type of feature information may refer to mapping the second type of feature information into feature vectors of a multidimensional space, so as to represent the operational behavior feature information of the user in different dimensions through a continuous feature vector, so that a value of the generated second feature vector mapped in a certain dimension represents a value corresponding to the feature information of the user in the dimension.
And then, the second feature vector and the first feature vector can be spliced to fuse static feature information and dynamic behavior feature information of the user to obtain a target feature vector after feature fusion, and the target feature vector can be used for representing a user portrait of the user.
In this way, when the object recommendation is performed, the target recommendation object can be determined by calculating the similarity between the target feature vector and the third feature vector corresponding to the candidate recommendation object.
The obtaining the second type of feature information of the user may include:
obtaining K pieces of operation object information of the user, wherein the K pieces of operation object information are information of K operation objects aimed at by the operation behaviors of the user, and K is an integer greater than or equal to 1;
respectively acquiring operation behavior information of the user on each operation object in the K operation objects to acquire M operation behavior information;
the generating the second feature vector corresponding to the second type of feature information may include:
respectively generating first sub-vectors corresponding to each piece of operation object information in the K pieces of operation object information to obtain K pieces of first sub-vectors;
generating first sub-vectors corresponding to each piece of operation behavior information in the M pieces of operation behavior information respectively to obtain M first sub-vectors;
respectively splicing first sub-vectors corresponding to the operation behavior information of the same operation object in the M first sub-vectors to obtain K second sub-vectors;
Respectively splicing the first sub-vectors and the second sub-vectors corresponding to the same operation object in the K first sub-vectors and the K second sub-vectors to obtain K third sub-vectors;
and splicing each third component in the K third component vectors to obtain the second feature vector corresponding to the second type feature information.
In this embodiment, for the second type of feature information, that is, the dynamic behavior feature information of the user, in order to generate a feature vector with finer granularity, the relationship between the operation behavior of the user and the operation objects (item) may be separated, and the operation behavior information of the user on different operation objects may be acquired based on the difference of the operation objects. For example, the playing operation behaviors of the user on different songs may be respectively obtained, specifically, the playing and collecting operation of the user on the song 1, the playing, collecting and sharing operation of the user on the song 2, the playing and dislike comment operation of the user on the song 3, and so on.
The obtained K pieces of operation object information are information of K operation objects for which the operation behavior of the user is directed, such as K different songs. The M pieces of operation behavior information are sets of operation behavior information of the obtained user on each operation object of the K operation objects, where, for each operation object, at least one piece of operation behavior information on the operation object may be obtained, that is, M is greater than K.
Then, each piece of operation object information in the K pieces of operation object information may be mapped to a corresponding vector, specifically, based on the attribute feature of each piece of operation object information, a respective corresponding first component vector may be generated, so as to obtain K first component vectors, for example, for a certain song, a corresponding first component vector may be generated based on the attribute features of singer, song wind, age, language, and the like of the song.
For each operation behavior information in the M operation behavior information, a corresponding first sub-vector may be generated respectively, so as to obtain M first sub-vectors, and based on the operation objects corresponding to the operation behavior information, the first sub-vectors corresponding to the operation behavior information of the same operation object may be spliced, so that for the K operation objects, the first sub-vectors corresponding to the operation behavior information of each operation object may be spliced, so as to fuse the operation behavior information belonging to the same operation object, so as to obtain K second sub-vectors. It should be noted that, the first sub-vectors of the operation behavior information corresponding to the same operation object may be in the same dimension, and the difference is only that the vector lengths are different, so that when the first sub-vectors corresponding to the same operation object are spliced, the values of the vectors may be added.
Then, for the K first sub-vectors and the K second sub-vectors, the first sub-vector and the second sub-vector corresponding to the same operation object may be spliced, so as to fuse each operation object and the operation behavior information corresponding to each operation object, and obtain K third sub-vectors.
For example, for a song, a corresponding first component vector may be generated, and for a playing operation and a collecting operation of the user on the song, a corresponding first sub-vector may be generated respectively, and the two first sub-vectors are spliced to obtain a corresponding second component vector, and then the generated first component vector and the second component vector may be spliced to obtain a third component vector, where the third component vector may represent a preference degree of the user on the song.
And finally, splicing each third vector in the K third sub-vectors to fuse the operation behavior information of all operation objects by the user to obtain the second feature vector capable of representing the dynamic operation behavior of the user.
Therefore, through the vector generation and splicing mode, the fusion of the operation behavior characteristics of the user can be realized, and different operation behaviors of the user on different operation objects can be finely grained, so that a user image which is finally constructed according to a first feature vector corresponding to the static feature information of the user and a second feature vector corresponding to the dynamic behavior feature information of the user has finer descriptive capability, and the recommendation effect of the object recommendation can be further ensured.
And 104, recommending the target recommended objects with the similarity meeting the preset condition to the user.
The similarity satisfies a preset condition, and the similarity may be that the similarity is greater than a preset threshold, such as that the similarity is greater than 75%, 80% or 85%, or the similarity is ranked in the top M bits, such as the top 3 bits, the top 2 bits, or the like.
In the embodiment of the invention, after the similarity between the first type characteristic information of the user and the candidate recommended object is obtained through calculation, a target recommended object, the similarity of which meets a preset condition, in the candidate recommended object can be determined, and the target recommended object is recommended to the user, namely, the target recommended object is recommended to a user account corresponding to the user.
For example, when it is determined that the feature vector corresponding to a certain song has a higher similarity with the feature vector corresponding to the user, it may be determined that the song is more in accordance with the listening preference of the user, so that the song may be recommended to the user.
In the embodiment of the present invention, the recommending apparatus may be any device having a storage medium, for example: computers (computers), cell phones, tablet computers (Tablet Personal Computer), laptops (Laptop computers), personal digital assistants (Personal Digital Assistant, PDA for short), mobile internet appliances (Mobile Internet Device, MID for short), or Wearable devices (weardabie devices), etc.
The following describes an embodiment of the present invention by way of example with reference to fig. 2:
for static characteristic information of a user, such as age, gender, region, occupation and the like, generating a corresponding first characteristic vector by adopting an embedding and table look-up (embedding) mode, and fusing all the static characteristic information by adopting a vector splicing mode to obtain fused user static characteristic information (user-information-embedding);
for the recent dynamic behavior characteristic information of the user, such as playing songs, collecting songs, sharing songs, clicking songs and the like, according to specific operation objects (items), and for the operation behaviors (actions) of each operation object, first sub-vectors of different operation objects are respectively generated in an embedded (webdding) manner, corresponding sub-vectors are respectively generated for the operation behaviors of different operation objects, the sub-vectors of the operation behaviors of each operation object are respectively spliced once, so as to obtain second sub-vectors of the operation behaviors of each operation object after fusion, then the second sub-vectors of the operation behaviors of each operation object and the second sub-vectors of the corresponding operation behaviors are further spliced twice, so as to obtain third sub-vectors of the operation behaviors of each operation object after fusion, and finally the third sub-vectors are spliced for three times, so as to obtain the recent dynamic behavior characteristic information (user-action-webdding) of the user, so as to obtain the recent characteristic vector capable of representing the user behavior.
Then, feature fusion can be carried out on the user static feature information and the user dynamic behavior feature information by splicing the first feature vector and the second feature vector, so as to obtain fused user feature information (user-embedding), namely, a target feature vector capable of representing the user portrait is obtained.
For the candidate recommended object, a corresponding third feature vector may be generated based on attribute features of the candidate recommended object, such as attribute features of singer, song wind, age, language, etc. of the song to be recommended, and a document embedding algorithm, such as doc2vec, may be adopted.
Finally, the recommendation of the related object can be performed based on the user portrait, specifically, the correlation between the user and the candidate recommended object can be calculated by calculating the cosine similarity between the third feature vector corresponding to the candidate recommended object and the target feature vector corresponding to the user portrait, so that the recommendation of the related object according to the user preference can be realized.
According to the recommendation method in the embodiment, the feature information of different dimensions of the user is mapped into continuous feature vectors, and the target recommendation objects with the similarity meeting the preset condition are recommended to the user based on the similarity comparison between the feature vectors of the user and the feature vectors of the candidate recommendation objects. Therefore, the user portraits represented by the continuous feature vectors do not fracture the correlation among the features of different dimensions of the user, so that when the object recommendation is performed based on the user feature vectors, the recommended object and the comprehensive features of the dimensions of the user can be ensured to be more matched, and further, a good recommendation effect can be ensured.
Referring to fig. 3, fig. 3 is a schematic structural diagram of a recommending apparatus according to an embodiment of the present invention, and as shown in fig. 3, a recommending apparatus 300 includes:
a first obtaining module 301, configured to obtain first type of feature information of a user, where the first type of feature information includes N pieces of feature information, N is an integer greater than 1;
a first generating module 302, configured to generate a first feature vector corresponding to the first type of feature information, where the first feature vector is an N-dimensional space vector, different dimensions in the N-dimensional space correspond to different feature information in the N feature information, and a user portrait of the user is represented by the first feature vector;
a calculating module 303, configured to calculate a similarity between the first feature vector and a third feature vector corresponding to the candidate recommendation object;
and the recommending module 304 is configured to recommend, to the user, a target recommended object whose similarity satisfies a preset condition.
Optionally, the first type of feature information is static feature information of the user;
the recommendation device 300 further includes:
the second acquisition module is used for acquiring second type characteristic information of the user, wherein the second type characteristic information comprises M pieces of characteristic information, M is an integer greater than 1, and the second type characteristic information is dynamic behavior characteristic information of the user;
The second generation module is used for generating a second feature vector corresponding to the second type of feature information;
the splicing module is used for splicing the second characteristic vector and the first characteristic vector to obtain a target characteristic vector, wherein the user portrait of the user is represented by the target characteristic vector;
the calculating module 303 is configured to calculate a similarity between the target feature vector and a third feature vector corresponding to the candidate recommendation object.
Optionally, the second obtaining module includes:
a first obtaining unit, configured to obtain K pieces of operation object information of the user, where the K pieces of operation object information are information of K operation objects aimed at by an operation behavior of the user, and K is an integer greater than or equal to 1;
the second acquisition unit is used for respectively acquiring the operation behavior information of the user on each operation object in the K operation objects to obtain M operation behavior information;
the second generation module includes:
the first generation unit is used for respectively generating first sub-vectors corresponding to each piece of operation object information in the K pieces of operation object information to obtain K pieces of first sub-vectors;
the second generating unit is used for respectively generating first sub-vectors corresponding to each piece of operation behavior information in the M pieces of operation behavior information to obtain M pieces of first sub-vectors;
The first splicing unit is used for respectively splicing the first sub-vectors corresponding to the operation behavior information of the same operation object in the M first sub-vectors to obtain K second sub-vectors;
the second splicing unit is used for respectively splicing the first sub-vectors and the second sub-vectors corresponding to the same operation object in the K first sub-vectors and the K second sub-vectors to obtain K third sub-vectors;
and the third splicing unit is used for splicing each third component in the K third component vectors to obtain the second characteristic vector corresponding to the second type characteristic information.
Optionally, the first generating module 402 includes:
the third generating unit is used for respectively generating second sub-vectors corresponding to each piece of characteristic information in the N pieces of characteristic information, and each second sub-vector corresponds to different dimensions;
and the fourth splicing unit is used for splicing each second sub-vector to obtain the first feature vector corresponding to the first type of feature information.
Optionally, the third generating unit includes:
the coding subunit is used for respectively coding each piece of characteristic information in the N pieces of characteristic information to obtain a coding sequence corresponding to each piece of characteristic information;
And the generation subunit is used for respectively generating second sub-vectors of corresponding dimensions from the coding sequences corresponding to the characteristic information by adopting a word embedding mode.
The recommending apparatus 300 is capable of implementing each process implemented by the recommending apparatus in the method embodiment of fig. 1, and in order to avoid repetition, a description thereof will be omitted. The recommendation device 300 according to the embodiment of the present invention may recommend the target recommended object whose similarity satisfies the preset condition to the user by mapping the feature information of different dimensions of the user into the continuous feature vector and by comparing the similarity between the feature vector of the user and the feature vector of the candidate recommended object. Therefore, the user portraits represented by the continuous feature vectors do not fracture the correlation among the features of different dimensions of the user, so that when the object recommendation is performed based on the user feature vectors, the recommended object and the comprehensive features of the dimensions of the user can be ensured to be more matched, and further, a good recommendation effect can be ensured.
The embodiment of the invention also provides a recommending device, which comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein the computer program realizes the processes of the recommending method embodiment when being executed by the processor, and can achieve the same technical effects, and the repetition is avoided, so that the description is omitted.
The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the processes of the above-mentioned preferred method embodiment, and can achieve the same technical effects, and in order to avoid repetition, the description is omitted here. Wherein the computer readable storage medium is selected from Read-Only Memory (ROM), random access Memory (Random Access Memory, RAM), magnetic disk or optical disk.
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, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The embodiments of the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present invention and the scope of the claims, which are to be protected by the present invention.
Claims (6)
1. A recommendation method, comprising:
acquiring first type characteristic information of a user, wherein the first type characteristic information comprises N pieces of characteristic information, and N is an integer greater than 1;
generating a first feature vector corresponding to the first type of feature information, wherein the first feature vector is an N-dimensional space vector, different dimensions in the N-dimensional space correspond to different feature information in the N feature information, and a user portrait of the user is represented by the first feature vector;
calculating the similarity between the first feature vector and a third feature vector corresponding to the candidate recommended object;
recommending target recommended objects with similarity meeting preset conditions in the candidate recommended objects to the user;
the first type of characteristic information is static characteristic information of the user; the method further comprises the steps of:
acquiring second type characteristic information of the user, wherein the second type characteristic information comprises M pieces of characteristic information, M is an integer greater than 1, and the second type characteristic information is dynamic behavior characteristic information of the user;
generating a second feature vector corresponding to the second type of feature information, wherein the second feature vector is a feature vector of a multidimensional space and represents the operation behavior feature information of the user in different dimensions;
Splicing the second characteristic vector and the first characteristic vector to obtain a target characteristic vector, wherein the user portrait of the user is represented by the target characteristic vector;
the calculating the similarity between the first feature vector and the third feature vector corresponding to the candidate recommended object includes:
calculating the similarity between the target feature vector and a third feature vector corresponding to the candidate recommended object;
the obtaining the second type characteristic information of the user comprises the following steps:
obtaining K pieces of operation object information of the user, wherein the K pieces of operation object information are information of K operation objects aimed at by the operation behaviors of the user, and K is an integer greater than or equal to 1;
respectively acquiring operation behavior information of the user on each operation object in the K operation objects to acquire M operation behavior information;
the generating the second feature vector corresponding to the second type of feature information includes:
respectively generating first sub-vectors corresponding to each piece of operation object information in the K pieces of operation object information to obtain K pieces of first sub-vectors;
generating first sub-vectors corresponding to each piece of operation behavior information in the M pieces of operation behavior information respectively to obtain M first sub-vectors;
Respectively splicing first sub-vectors corresponding to the operation behavior information of the same operation object in the M first sub-vectors to obtain K second sub-vectors;
respectively splicing the first sub-vectors and the second sub-vectors corresponding to the same operation object in the K first sub-vectors and the K second sub-vectors to obtain K third sub-vectors;
and splicing each third component in the K third component vectors to obtain the second feature vector corresponding to the second type feature information.
2. The method of claim 1, wherein generating the first feature vector corresponding to the first type of feature information comprises:
generating second sub-vectors corresponding to each piece of characteristic information in the N pieces of characteristic information respectively, wherein each second sub-vector corresponds to different dimensions respectively;
and splicing each second sub-vector to obtain the first feature vector corresponding to the first type of feature information.
3. The method according to claim 2, wherein generating the second sub-vector corresponding to each of the N pieces of feature information includes:
respectively encoding each piece of characteristic information in the N pieces of characteristic information to obtain a coding sequence corresponding to each piece of characteristic information;
And generating second sub-vectors of corresponding dimensions from the coding sequences corresponding to each piece of characteristic information by adopting a word embedding mode.
4. A recommendation device, comprising:
the device comprises a first acquisition module, a second acquisition module and a first processing module, wherein the first acquisition module is used for acquiring first type characteristic information of a user, the first type characteristic information comprises N pieces of characteristic information, and N is an integer larger than 1;
the first generation module is used for generating a first feature vector corresponding to the first type of feature information, wherein the first feature vector is an N-dimensional space vector, different dimensions in the N-dimensional space correspond to different feature information in the N feature information, and a user portrait of the user is represented through the first feature vector;
the computing module is used for computing the similarity between the first feature vector and a third feature vector corresponding to the candidate recommended object;
the recommendation module is used for recommending target recommended objects, of which the similarity meets a preset condition, to the user;
the first type of characteristic information is static characteristic information of the user; the recommendation device further includes:
the second acquisition module is used for acquiring second type characteristic information of the user, wherein the second type characteristic information comprises M pieces of characteristic information, M is an integer greater than 1, and the second type characteristic information is dynamic behavior characteristic information of the user;
The second generation module is used for generating a second feature vector corresponding to the second type of feature information, wherein the second feature vector is a feature vector of a multidimensional space and represents the operation behavior feature information of the user in different dimensions;
the splicing module is used for splicing the second characteristic vector and the first characteristic vector to obtain a target characteristic vector, wherein the user portrait of the user is represented by the target characteristic vector;
the calculation module is used for calculating the similarity between the target feature vector and a third feature vector corresponding to the candidate recommended object;
the second acquisition module includes:
a first obtaining unit, configured to obtain K pieces of operation object information of the user, where the K pieces of operation object information are information of K operation objects aimed at by an operation behavior of the user, and K is an integer greater than or equal to 1;
the second acquisition unit is used for respectively acquiring the operation behavior information of the user on each operation object in the K operation objects to obtain M operation behavior information;
the second generation module includes:
the first generation unit is used for respectively generating first sub-vectors corresponding to each piece of operation object information in the K pieces of operation object information to obtain K pieces of first sub-vectors;
The second generating unit is used for respectively generating first sub-vectors corresponding to each piece of operation behavior information in the M pieces of operation behavior information to obtain M pieces of first sub-vectors;
the first splicing unit is used for respectively splicing the first sub-vectors corresponding to the operation behavior information of the same operation object in the M first sub-vectors to obtain K second sub-vectors;
the second splicing unit is used for respectively splicing the first sub-vectors and the second sub-vectors corresponding to the same operation object in the K first sub-vectors and the K second sub-vectors to obtain K third sub-vectors;
and the third splicing unit is used for splicing each third component in the K third component vectors to obtain the second characteristic vector corresponding to the second type characteristic information.
5. A recommendation device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the recommendation method according to any one of claims 1 to 3.
6. A computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps in the recommendation method according to any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010289694.0A CN111488526B (en) | 2020-04-14 | 2020-04-14 | Recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010289694.0A CN111488526B (en) | 2020-04-14 | 2020-04-14 | Recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111488526A CN111488526A (en) | 2020-08-04 |
| CN111488526B true CN111488526B (en) | 2024-04-05 |
Family
ID=71812755
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010289694.0A Active CN111488526B (en) | 2020-04-14 | 2020-04-14 | Recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111488526B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112416931A (en) * | 2020-11-18 | 2021-02-26 | 脸萌有限公司 | Information generation method and device and electronic equipment |
| CN113157898B (en) * | 2021-05-26 | 2022-10-14 | 中国平安人寿保险股份有限公司 | Method and device for recommending candidate questions, computer equipment and storage medium |
| CN113254804B (en) * | 2021-07-06 | 2021-12-03 | 武汉荟友网络科技有限公司 | Social relationship recommendation method and system based on user attributes and behavior characteristics |
| CN117495515B (en) * | 2023-12-29 | 2024-04-05 | 优材优建(青岛)供应链科技有限公司 | Bid intelligent matching method, system, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014215685A (en) * | 2013-04-23 | 2014-11-17 | 株式会社Ｎｔｔドコモ | Recommendation server and recommendation content determination method |
| CN110232152A (en) * | 2019-05-27 | 2019-09-13 | 腾讯科技（深圳）有限公司 | Content recommendation method, device, server and storage medium |
| CN110275980A (en) * | 2019-06-26 | 2019-09-24 | 徐州工业职业技术学院 | One kind having an X-rayed music recommended method based on group |
| CN110866181A (en) * | 2019-10-12 | 2020-03-06 | 平安国际智慧城市科技股份有限公司 | Resource recommendation method, device and storage medium |
-
2020
- 2020-04-14 CN CN202010289694.0A patent/CN111488526B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014215685A (en) * | 2013-04-23 | 2014-11-17 | 株式会社Ｎｔｔドコモ | Recommendation server and recommendation content determination method |
| CN110232152A (en) * | 2019-05-27 | 2019-09-13 | 腾讯科技（深圳）有限公司 | Content recommendation method, device, server and storage medium |
| CN110275980A (en) * | 2019-06-26 | 2019-09-24 | 徐州工业职业技术学院 | One kind having an X-rayed music recommended method based on group |
| CN110866181A (en) * | 2019-10-12 | 2020-03-06 | 平安国际智慧城市科技股份有限公司 | Resource recommendation method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111488526A (en) | 2020-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111488526B (en) | Recommendation method and device | |
| CN109002488B (en) | Recommendation model training method and device based on meta-path context | |
| Jalili et al. | Evaluating collaborative filtering recommender algorithms: a survey | |
| CN106776673B (en) | Multimedia document summarization | |
| CN111079028B (en) | Collaborative filtering recommendation system and method based on multi-source auxiliary information | |
| US10558852B2 (en) | Predictive analysis of target behaviors utilizing RNN-based user embeddings | |
| CN108170684B (en) | Text similarity calculation method and system, data query system and computer product | |
| Sieg et al. | Improving the effectiveness of collaborative recommendation with ontology-based user profiles | |
| Gedikli et al. | Improving recommendation accuracy based on item-specific tag preferences | |
| Lai et al. | Novel personal and group-based trust models in collaborative filtering for document recommendation | |
| JP2005317018A (en) | Method and system for calculating importance of block in display page | |
| CN111522889B (en) | User interest tag expansion method and device, electronic equipment and storage medium | |
| TW201814556A (en) | Information matching method and related device | |
| Angadi et al. | Multimodal sentiment analysis using reliefF feature selection and random forest classifier | |
| da Silva et al. | Content-based social recommendation with poisson matrix factorization | |
| Liu et al. | Towards context-aware collaborative filtering by learning context-aware latent representations | |
| Li et al. | Heterogeneous graph embedding for cross-domain recommendation through adversarial learning | |
| CN111291563A (en) | Word vector alignment method and training method of word vector alignment model | |
| Fareed et al. | A collaborative filtering recommendation framework utilizing social networks | |
| CN114065016A (en) | Recommendation method, device, equipment and computer readable storage medium | |
| Dakhel et al. | Providing an effective collaborative filtering algorithm based on distance measures and neighbors’ voting | |
| CN112100507B (en) | Object recommendation method, computing device and computer-readable storage medium | |
| Wang et al. | Park recommendation algorithm based on user reviews and ratings | |
| CN113641915A (en) | Object recommendation method, device, equipment, storage medium and program product | |
| CN113887613A (en) | Deep learning method, device and equipment based on attention mechanism and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |