CN110941758B - Synthetic feature generation method and device of recommendation system - Google Patents
Synthetic feature generation method and device of recommendation system Download PDFInfo
- Publication number
- CN110941758B CN110941758B CN201911113893.XA CN201911113893A CN110941758B CN 110941758 B CN110941758 B CN 110941758B CN 201911113893 A CN201911113893 A CN 201911113893A CN 110941758 B CN110941758 B CN 110941758B
- Authority
- CN
- China
- Prior art keywords
- matrix
- dimensional
- row
- feature
- history
- 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
Images
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present specification provides a method for generating a composite feature of a recommendation system, including: generating an n multiplied by m dimensional history and object matrix P based on n history characteristics and m target object characteristics of a certain user; n and m are natural numbers; generating an m-dimensional intermediate vector theta according to all row vectors of the row normalization matrix B, and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A; the row normalization matrix B is an n x m dimensional matrix obtained by normalizing each row of the history and object matrix P, and the column normalization matrix A is an n x m dimensional matrix obtained by normalizing each column of the history and object matrix P; generating a composite feature from the weight vector gamma and the n historical features; and inputting the synthesized features into a machine learning recommendation model, and predicting the matching degree of the user and the target object.
Description
Technical Field
The present disclosure relates to the field of data processing technologies, and in particular, to a method and an apparatus for generating a synthetic feature of a recommendation system.
Background
One important application of data mining is to make predictions for the future using historical data. For example, tomorrow's weather is predicted from weather data over the past years, new movies that the user may be interested in are predicted from movies he has seen before, and so on. Some recommendation systems use features describing a predicted subject (such as features of the user himself or herself and features of a movie that the user has watched before) and features describing a predicted target object (such as features of a new movie) as input features, and use a machine learning model as a recommendation algorithm to obtain a matching degree between the predicted subject and the target object (such as a degree of interest of the user in the new movie).
In order to improve the accuracy of the recommendation system, not only input features of the recommendation system (features describing the prediction subject and features describing the target object) may be used as inputs of the machine learning model, but also synthesized features derived from the input features of the recommendation system may be used as inputs of the machine learning model. The more the synthetic features can embody the correlation between the input features of the recommendation system, the more the prediction accuracy can be improved. However, the cost of introducing the synthetic features is the increase of the input dimension of the machine learning model, the higher the dimension of the synthetic features is, the more the parameters of the machine learning model are, the higher the difficulty of training the machine learning model is, and the larger the calculation resources required by training are.
Disclosure of Invention
In view of this, the present specification provides a method for generating a synthetic feature of a recommendation system, including:
generating an n multiplied by m dimensional history and object matrix P based on n history characteristics and m target object characteristics of a certain user; the above-mentioned
Wherein h is u Is an n x d dimensional matrix composed of n d dimensional historical feature vectors, h a An m x d dimensional matrix formed by m d dimensional target object feature vectors; d. n and m are natural numbers;
generating an m-dimensional intermediate vector theta according to all row vectors of the row normalization matrix B, and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A; the row normalization matrix B is an n x m dimensional matrix obtained by normalizing each row of the history and object matrix P, and the column normalization matrix A is an n x m dimensional matrix obtained by normalizing each column of the history and object matrix P;
generating a composite feature from the weight vector γ and the n historical features;
and inputting the synthesized features into a machine learning recommendation model, and predicting the matching degree of the user and the target object.
The present specification also provides a synthetic feature generation apparatus of a recommendation system, including:
a history and object matrix unit, which is used for generating a history and object matrix P with dimension of n multiplied by m based on n history characteristics and m target object characteristics of a user; the above-mentioned
Wherein h is u Is an n x d dimensional matrix composed of n d dimensional historical feature vectors, h a An m x d dimensional matrix formed by m d dimensional target object feature vectors; d. n and m are natural numbers;
the weight vector unit is used for generating an m-dimensional intermediate vector theta according to all row vectors of the row normalization matrix B and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A; the row normalization matrix B is an n x m dimensional matrix obtained after normalization processing is carried out on each row of the history and object matrix P, and the column normalization matrix A is an n x m dimensional matrix obtained after normalization processing is carried out on each column of the history and object matrix P;
a composite feature unit for generating a composite feature from the weight vector γ and the n history features;
and the characteristic output unit is used for inputting the synthesized characteristic into a machine learning recommendation model and predicting the matching degree of the user and the target object.
This specification provides a computer device comprising: a memory and a processor; the memory having stored thereon a computer program executable by the processor; and when the processor runs the computer program, executing the method of the synthetic feature generation method of the recommendation system.
The present specification also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the method of the synthetic feature generation method of the recommendation system described above.
According to the technical scheme, in the embodiment of the specification, a history and object matrix P is generated based on n user history characteristics and m target object characteristics input into a recommendation system, a weight vector gamma is obtained from a row normalization matrix B and a column normalization matrix A of the P, and a synthesized characteristic is generated through the weight vector gamma and the n history characteristics; the dimension of the synthesized feature is only equivalent to the dimension of one historical feature, the influence on the training difficulty of the machine learning prediction model is extremely limited, and the computing resources are saved.
Drawings
FIG. 1 is a schematic structural diagram of a recommendation system in an embodiment of the present disclosure;
FIG. 2 is a flow chart of a method for generating synthetic features of a recommendation system in an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a calculation process for generating a synthesized feature in an application example of the present specification;
FIG. 4 is a hardware block diagram of an apparatus for performing embodiments of the present description;
fig. 5 is a logical block diagram of a synthesized feature generation apparatus of a recommendation system in an embodiment of the present specification.
Detailed Description
The embodiment of the specification provides a new synthetic feature generation method of a recommendation system, which adopts n user history features and m target object features to generate an n x m-dimensional history and object matrix P, respectively performs row and column normalization processing on the matrix P, obtains a weight vector gamma through a column normalization matrix A and a row normalization matrix B, and obtains synthetic features by applying the weight vector gamma to the n history features, so that the synthetic features generated based on a double-layer attention mechanism can more deeply dig out the association between the history features and the target object features, and the prediction accuracy can be improved; because the dimension of the synthesized feature is equivalent to a historical feature, the quantity of parameters of the machine learning prediction model is greatly limited, and the computing resources consumed for training the machine learning recommendation model can be saved.
In the embodiments of the present description, the recommendation system is used to predict the matching degree between a certain user and several target objects. Wherein the target object may be an alternative object intended to be recommended to the user, e.g. a recommendation system recommending movies to the user, the target object may be a movie being shown. The recommendation system in the embodiment of the present specification may have a structure shown in fig. 1. The input features of the recommendation system include at least n (n is a natural number) historical features of a certain user, and at least m (m is a natural number) target object features, and may also include other features, without limitation. The n history features and the m target object features in the input features are used as the input of the composite feature function module. And the synthetic feature function module runs the embodiment of the specification and outputs the synthetic features to the machine learning recommendation model in the recommendation system. The synthesized features are typically used as input to a machine learning recommendation model that outputs a degree of matching of the user to the target object, along with input features of the recommendation system. The embodiment of the present specification does not limit the algorithm used by the machine learning recommendation model, such as a regression algorithm, a decision tree algorithm, a limit boltzmann machine algorithm, and the like.
The historical characteristics of the user are generated according to the historical behavior record of the user, and may be descriptions of the historical behavior itself, or descriptions of objects related to the historical behavior (such as targets targeted by the historical behavior), or descriptions of the historical behavior itself and the objects related to the historical behavior; the target object characteristic is the description of the target object and the related information thereof; are not limited. For example, in a recommendation system for recommending a movie to a user, the historical characteristics may include information such as a category in which the user watched the movie, a movie watching time period of the user, and an area in which the user went to a movie theater; the target object characteristics may include information on the category of the movie being shown, the time period for showing, the area where the movie theater is shown, and the like. In addition, the historical behavior record of the user can include one or more historical behaviors according to the requirements of the actual application scene, and is not limited as well; for example, in a recommendation system for recommending commodities to a user, the history characteristics may include a characteristic that the user purchased a commodity, a characteristic that the user browsed a commodity, a characteristic that the user collected a commodity, and the like.
Embodiments of the present description may be implemented on any device with computing and storage capabilities, such as a mobile phone, a tablet Computer, a PC (Personal Computer), a notebook, a server, and so on; the functions in the embodiments of the present specification may also be implemented by a logical node operating in two or more devices.
In the embodiment of the present specification, a flow of a synthetic feature generation method of a recommendation system is shown in fig. 2.
In data mining, a feature is usually expressed as a vector, so that n historical features can be expressed as a historical feature matrix, and m target object features can be expressed as a target object feature matrix. In the embodiments of the present description, according to factors such as the historical characteristics and the target object characteristics in an actual application scenario, and the incidence relation, various matrix operation methods may be adopted, and various variables or constants may be introduced to generate the n × m-dimensional historical and object matrix P from the historical characteristic matrix and the target object matrix, without limitation.
In some application scenarios, the characteristics of the user historical behavior object (i.e. the object recommended by the recommendation system involved in the user historical behavior) are taken as the historical characteristics. For example, in a recommendation system that recommends a product to a user, a user feature is a feature of a product that the user has purchased, a product that the user has viewed, or a product that the user has collected, and each user feature is a product feature, and a target object recommended to the user is also a product, and a target object feature is also a product feature. It can be seen that in these application scenarios, the history feature is a feature of an object recommended by the recommendation system, and the target object feature is also a feature of an object recommended by the recommendation system, so the history feature vector and the target object feature vector have the same dimension.
In the application scenario in which the user historical behavior object is taken as the historical characteristic, the dimension of the historical characteristic vector and the dimension of the target object characteristic vector are set to be d, and a historical characteristic matrix h formed by n d-dimensional historical characteristic vectors is set u A target object feature matrix h composed of n x d dimensional matrix and m d dimensional target object feature vectors a For an m × d matrix, the history and object matrix P can be obtained by equation 1:
step 220: and generating an m-dimensional intermediate vector theta according to all the row vectors of the row normalization matrix B, and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A.
The column normalization matrix a is an n × m-dimensional matrix obtained by normalizing each column of the history and object matrix P, and the row normalization matrix B is an n × m-dimensional matrix obtained by normalizing each row of the history and object matrix P.
Let the element of the history and object matrix P be P ij Then the column normalizes the element α of the matrix A ij The element beta of the row normalization matrix B can be derived from equation 2 ij This can be derived from equation 3:
the row vectors of the row normalization matrix B are m-dimensional and n in number. The specific manner of calculating the m-dimensional intermediate vector θ from the n m-dimensional row vectors and the specific manner of calculating the n-dimensional weight vector γ from the intermediate vector and the column normalization matrix a may be determined according to the needs of the actual application scenario, and the embodiments of the present specification are not limited. For example, the row vectors of the row normalization matrix B may be summed, weighted, and the like to obtain the intermediate vector θ.
In one implementation, the average of all row vectors of the row normalization matrix B may be taken as the intermediate vector θ, i.e., the element θ of the intermediate vector θ j This can be derived from equation 4:
in another implementation, the column normalization matrix a and the intermediate vector θ may be dot-multiplied to obtain an n-dimensional weight vector γ, that is, the weight vector γ may be obtained by equation 5, and the element γ of the weight vector γ is i This can be derived from equation 6:
γ＝A·θ T formula 5
Step 230: from the weight vector γ and the n historical features, a composite feature is generated.
The element y of the weight vector y can be expressed as i And the weighted sum of the n historical characteristics and the weight value of the ith historical characteristic corresponding to the element is obtained to obtain the synthesized characteristic.
Specifically, the n-dimensional weight vector γ and the historical feature matrix h can be combined u Performing dot multiplication to obtain a d-dimensional composite feature vector s, i.e. assuming that the composite feature vector s can be obtained by equation 7:
s＝γ·h u formula 7
It can be seen that, in the process of generating the synthesized features, the history and object matrix P reflects the user's reflection on different target objects, a first layer of attention mechanism is introduced when the column normalization matrix a and the row normalization matrix B of the history and object matrix P are calculated, and a second layer of attention mechanism is introduced when the weight vector γ is generated by the column normalization matrix a and the row normalization matrix B. Therefore, the synthetic features s can not only express the difference of different target objects, but also extract interest transition of the user in a time dimension based on a double-layer attention mechanism, so that the association between the user features and the target object features is better. In addition, the d-dimensional synthetic feature is only equivalent to the dimension of one user feature, the influence on the input dimension of the machine learning recommendation model is almost negligible, and the parameter overhead of the machine learning recommendation model is basically not increased.
Step 240: and inputting the synthesized features into a machine learning recommendation model, and predicting the matching degree of the user and the target object.
The generated composite features may be input to a machine learning recommendation model within the recommendation system, which uses the composite features to predict how well the user matches the target object. In addition to the synthesized features, other input features of the recommendation system may be used as input to the machine-learned recommendation model, such as one or more of historical features, target object features, and other features of the user; in addition, other methods besides the embodiments of the present specification may be adopted, and other synthetic features may be generated from the input features of the recommendation system as the input of the machine learning recommendation model.
It can be seen that in the embodiment of the present specification, n user history features and m target object features are adopted to generate a history and object matrix P, a weight vector γ is obtained from a row normalization matrix B and a column normalization matrix a of P, and a composite feature is obtained by applying the weight vector γ to the n history features, so that the composite feature under a double-layer attention mechanism can better reflect the association between a user and a target object, and the accuracy of a recommendation system is improved; and the dimension of the synthesized feature is equivalent to a historical feature, so that the parameter number of the machine learning prediction model is greatly limited, the training difficulty of the machine learning recommendation model can be reduced, and the computing resources required by training are saved.
The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
In one application example of the present specification, an internet service provider that employs a recommendation system to determine which advertisement item is to be recommended to a user among m advertisement items (a target object) pushes the advertisement item to the user.
The input features of the recommendation system are: portrait characteristics of the user, n history characteristics of the user, and m advertisement goods characteristics (a kind of target object characteristics). Wherein, the portrait characteristics of the user carry the information of the user such as age, sex, income grade, consumption grade and the like; the historical characteristics of the user are extracted from the historical behavior record of the user, and specifically include: a feature that the user purchased the merchandise, a feature that the user browsed the merchandise, or a feature that the user collected the merchandise. Both the user's historical characteristics and the advertised item characteristics are characteristics of the item, having the same dimension d. The characteristics of the commodity carry information such as the category of the commodity, the attribute of the commodity, the picture of the commodity and the like.
The recommendation system generates a composite feature according to the flow illustrated in fig. 3, using the n historical features of the user (i.e., the product features extracted from the n historical behavior records of the user) and the m advertisement product features. In fig. 3, n is 5, m is 2, and d is 3.
N historical characteristics of the user form a historical characteristic matrix h u ，h u ∈R n×d (ii) a m advertisement commodity characteristics form a target object characteristic matrix h a ，h a ∈R m×d 。
From the historical feature matrix h, according to equation 1 u And a target object feature matrix h a The point multiplication of (A) to (D) yields a history and object matrix P, P ∈ R n×m 。
Softmax (normalization) is respectively carried out on the history matrix P and the object matrix P in the row direction and the column direction, and a column normalization matrix A and a row normalization matrix B are obtained according to an equation 2 and an equation 3. This is a calculation of the first layer attention mechanism. Wherein A ∈ R n×m ，B∈R n×m 。
The rows of the row normalization matrix B are averaged, and the average of all row vectors of B is taken as the intermediate vector θ according to equation 4. And performing point multiplication on the column normalization matrix A and the intermediate vector theta according to the formula 5 and the formula 6 to obtain an n-dimensional weight vector gamma. This is a calculation of the second layer attention mechanism.
The weight vector gamma and the historical feature matrix h are combined u And performing dot multiplication on the time dimension to obtain a synthetic characteristic s according to the formula 7.
And inputting the synthesized features s into a subsequent machine learning recommendation model in the recommendation system. The machine learning recommendation model predicts the advertisement commodity which is most interesting to the user (namely, is most matched with the user) according to the input composite feature s, the n history features of the user, the m advertisement commodity features and the portrait features of the user, and outputs the advertisement commodity as a recommendation system.
Corresponding to the above flow implementation, an embodiment of the present specification further provides a synthetic feature generation device of a recommendation system. The apparatus may be implemented by software, or by hardware, or by a combination of hardware and software. Taking a software implementation as an example, the logical means is formed by reading a corresponding computer program instruction into a memory by a Central Processing Unit (CPU) of the device in which the CPU is located. In terms of hardware, the device in which the synthesized feature generating apparatus of the recommended system is located generally includes other hardware such as a chip for transmitting and receiving wireless signals and/or other hardware such as a board for realizing a network communication function, in addition to the CPU, the memory, and the storage shown in fig. 4.
Fig. 5 is a synthesized feature generation apparatus of a recommendation system according to an embodiment of the present disclosure, including a history and object matrix unit, a weight vector unit, a synthesized feature unit, and a feature output unit, where: the history and object matrix unit is used for generating an n multiplied by m-dimensional history and object matrix P based on n history characteristics and m target object characteristics of a certain user; n and m are natural numbers; the weight vector unit is used for generating an m-dimensional intermediate vector theta according to all row vectors of the row normalization matrix B and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A; the row normalization matrix B is an n x m dimensional matrix obtained after normalization processing is carried out on each row of the history and object matrix P, and the column normalization matrix A is an n x m dimensional matrix obtained after normalization processing is carried out on each column of the history and object matrix P; the synthetic feature unit is used for generating synthetic features by the weight vector gamma and the n historical features; and the characteristic output unit is used for inputting the synthesized characteristic into a machine learning recommendation model and predicting the matching degree of the user and the target object.
Optionally, the generating, by the weight vector unit, an m-dimensional intermediate vector θ according to all row vectors of the row normalization matrix B includes: the average value of all row vectors of the row normalization matrix B is taken as the m-dimensional intermediate vector θ.
Optionally, the weight vector unit obtains an n-dimensional weight vector γ based on the intermediate vector θ and the column normalization matrix a, and includes: and performing point multiplication on the column normalization matrix A and the intermediate vector theta to obtain an n-dimensional weight vector gamma.
Optionally, the history and object matrix
Wherein h is u Is an n x d dimensional matrix composed of n d dimensional historical feature vectors, h a An m x d dimensional matrix formed by m d dimensional target object feature vectors; d is a natural number; the synthetic feature unit is specifically configured to: combining the n-dimensional weight vector gamma and the historical feature matrix h u And performing dot multiplication to obtain the d-dimensional synthetic feature.
Optionally, the history characteristics include: a characteristic that the user purchased the merchandise, a characteristic that the user browsed the merchandise, or a characteristic that the user collected the merchandise; the target object features include: characteristics of the advertised item.
Embodiments of the present description provide a computer device that includes a memory and a processor. Wherein the memory has stored thereon a computer program executable by the processor; the processor, when executing the stored computer program, performs the steps of the method of generating a composite feature of the recommendation system in the embodiments of the present specification. For a detailed description of the steps of the method for generating the synthetic features of the recommendation system, reference is made to the preceding contents, which are not repeated.
Embodiments of the present description provide a computer-readable storage medium having stored thereon computer programs which, when executed by a processor, perform the steps of the composite feature generation method of the recommendation system in embodiments of the present description. For a detailed description of the steps of the method for generating the synthetic features of the recommendation system, reference is made to the preceding contents, which are not repeated.
While the invention has been described in terms of what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications, equivalents, and alternatives falling within the spirit and scope of the invention as defined by the appended claims.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, embodiments of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.
Claims (12)
1. A method for generating synthetic features of a recommendation system includes:
generating an n multiplied by m dimensional history and object matrix P based on n history characteristics and m target object characteristics of a certain user; the above-mentioned
Wherein h is u Is an n x d dimensional matrix composed of n d dimensional historical feature vectors, h a An m x d dimensional matrix formed by m d dimensional target object feature vectors; d. n and m are natural numbers;
generating an m-dimensional intermediate vector theta according to all row vectors of the row normalization matrix B, and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A; the row normalization matrix B is an n x m dimensional matrix obtained after normalization processing is carried out on each row of the history and object matrix P, and the column normalization matrix A is an n x m dimensional matrix obtained after normalization processing is carried out on each column of the history and object matrix P;
generating a composite feature from the weight vector gamma and the n historical features;
and inputting the synthesized features into a machine learning recommendation model, and predicting the matching degree of the user and the target object.
2. The method of claim 1, the generating an m-dimensional intermediate vector θ from all row vectors of a row normalized matrix B, comprising: the average value of all row vectors of the row normalization matrix B is taken as the m-dimensional intermediate vector θ.
3. The method of claim 1, the deriving an n-dimensional weight vector γ based on an intermediate vector θ and a column normalization matrix a, comprising: and performing point multiplication on the column normalization matrix A and the intermediate vector theta to obtain an n-dimensional weight vector gamma.
4. The method of claim 1, the generating a composite feature from the weight vector γ and the n historical features, comprising: combining the n-dimensional weight vector gamma and the historical feature matrix h u And performing dot multiplication to obtain the d-dimensional synthetic feature.
5. The method of claim 1, the historical features comprising: a characteristic that the user purchased the merchandise, a characteristic that the user browsed the merchandise, or a characteristic that the user collected the merchandise; the target object features include: characteristics of the advertised item.
6. A composite feature generation apparatus of a recommendation system, comprising:
a history and object matrix unit, which is used for generating a history and object matrix P with dimension of n multiplied by m based on n history characteristics and m target object characteristics of a user; the above-mentioned
Wherein h is u Is an n x d dimensional matrix composed of n d dimensional historical feature vectors, h a An m x d dimensional matrix formed by m d dimensional target object feature vectors; d. n and m are natural numbers;
the weight vector unit is used for generating an m-dimensional intermediate vector theta according to all row vectors of the row normalization matrix B and obtaining an n-dimensional weight vector gamma based on the intermediate vector theta and the column normalization matrix A; the row normalization matrix B is an n x m dimensional matrix obtained after normalization processing is carried out on each row of the history and object matrix P, and the column normalization matrix A is an n x m dimensional matrix obtained after normalization processing is carried out on each column of the history and object matrix P;
a synthesized feature unit for generating a synthesized feature from the weight vector γ and the n history features;
and the characteristic output unit is used for inputting the synthesized characteristic into a machine learning recommendation model and predicting the matching degree of the user and the target object.
7. The apparatus of claim 6, the weight vector unit to generate an m-dimensional intermediate vector θ from all row vectors of a row normalization matrix B, comprising: the average value of all row vectors of the row normalization matrix B is taken as the m-dimensional intermediate vector θ.
8. The apparatus of claim 6, the weight vector unit to derive an n-dimensional weight vector γ based on an intermediate vector θ and a column normalization matrix A, comprising: and performing point multiplication on the column normalization matrix A and the intermediate vector theta to obtain an n-dimensional weight vector gamma.
9. The apparatus of claim 6, the synthesized feature unit to be specifically configured to: combining the n-dimensional weight vector gamma and the historical feature matrix h u And performing dot multiplication to obtain the d-dimensional synthetic feature.
10. The apparatus of claim 6, the historical features comprising: a characteristic that the user purchased the merchandise, a characteristic that the user browsed the merchandise, or a characteristic that the user collected the merchandise; the target object features include: characteristics of the advertised item.
11. A computer device, comprising: a memory and a processor; the memory having stored thereon a computer program executable by the processor; the processor, when executing the computer program, performs the method of any of claims 1 to 5.
12. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911113893.XA CN110941758B (en) | 2019-11-14 | 2019-11-14 | Synthetic feature generation method and device of recommendation system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911113893.XA CN110941758B (en) | 2019-11-14 | 2019-11-14 | Synthetic feature generation method and device of recommendation system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110941758A CN110941758A (en) | 2020-03-31 |
| CN110941758B true CN110941758B (en) | 2022-09-16 |
Family
ID=69906538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911113893.XA Active CN110941758B (en) | 2019-11-14 | 2019-11-14 | Synthetic feature generation method and device of recommendation system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110941758B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108711075A (en) * | 2018-05-22 | 2018-10-26 | 阿里巴巴集团控股有限公司 | A kind of Products Show method and apparatus |
| CN109214882A (en) * | 2018-07-09 | 2019-01-15 | 西北大学 | A kind of Method of Commodity Recommendation |
| CN110012356A (en) * | 2019-04-16 | 2019-07-12 | 腾讯科技（深圳）有限公司 | Video recommendation method, device and equipment and computer storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140181121A1 (en) * | 2012-12-21 | 2014-06-26 | Microsoft Corporation | Feature embedding in matrix factorization |
-
2019
- 2019-11-14 CN CN201911113893.XA patent/CN110941758B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108711075A (en) * | 2018-05-22 | 2018-10-26 | 阿里巴巴集团控股有限公司 | A kind of Products Show method and apparatus |
| CN109214882A (en) * | 2018-07-09 | 2019-01-15 | 西北大学 | A kind of Method of Commodity Recommendation |
| CN110012356A (en) * | 2019-04-16 | 2019-07-12 | 腾讯科技（深圳）有限公司 | Video recommendation method, device and equipment and computer storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于Spark的音乐推荐系统的设计与实现;顾威;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20190115;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110941758A (en) | 2020-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210027160A1 (en) | End-to-end deep collaborative filtering | |
| CN111626832B (en) | Product recommendation method and device and computer equipment | |
| CN112487278A (en) | Training method of recommendation model, and method and device for predicting selection probability | |
| US20190279231A1 (en) | Analytical precursor mining for personalized recommendation | |
| US11640617B2 (en) | Metric forecasting employing a similarity determination in a digital medium environment | |
| US11928616B2 (en) | Method and system for hierarchical forecasting | |
| US20200167690A1 (en) | Multi-task Equidistant Embedding | |
| US9269049B2 (en) | Methods, apparatus, and systems for using a reduced attribute vector of panel data to determine an attribute of a user | |
| CN110008397B (en) | Recommendation model training method and device | |
| CN111008335B (en) | Information processing method, device, equipment and storage medium | |
| US20220172083A1 (en) | Noise contrastive estimation for collaborative filtering | |
| US11687839B2 (en) | System and method for generating and optimizing artificial intelligence models | |
| CN114117216A (en) | Recommendation probability prediction method and device, computer storage medium and electronic equipment | |
| WO2023087914A1 (en) | Method and apparatus for selecting recommended content, and device, storage medium and program product | |
| KR20210144330A (en) | Method and apparatus for recommending item based on collaborative filtering neural network | |
| CN113269232B (en) | Model training method, vectorization recall method, related equipment and storage medium | |
| CN111161009B (en) | Information pushing method, device, computer equipment and storage medium | |
| CN110941758B (en) | Synthetic feature generation method and device of recommendation system | |
| US11269974B1 (en) | Learning ordinal regression model via divide-and-conquer technique | |
| WO2017095371A1 (en) | Product recommendations based on selected user and product attributes | |
| CN111768218A (en) | Method and device for processing user interaction information | |
| CN113378071A (en) | Advertisement recommendation method and device, electronic equipment and storage medium | |
| CN113792952A (en) | Method and apparatus for generating a model | |
| CN113822734A (en) | Method and apparatus for generating information | |
| CN112749005A (en) | Resource data processing method and device, computer equipment 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 |