Disclosure of Invention
The inventor finds through research that in the personalized recommendation modeling process, future behaviors are only predicted according to past behaviors of a user, and specific analysis is not performed on the behaviors. For example, some click actions of the user are false clicks; or the user clicks certain commodities, but the user finds that the commodities are disliked after browsing, so that the corresponding purchasing behavior is not available; or the user purchases a commodity for emergency, but will not typically purchase. Since no specific analysis is made of the user's behavior, the true intent of the user's behavior cannot be understood from the user's historical behavior alone.
To this end, the present disclosure provides a solution for recommending items to a user based on the user's attention.
According to one aspect of one or more embodiments of the present disclosure, there is provided an item recommendation method comprising: extracting a predetermined number of candidate items from an item pool, extracting characteristics of a user history behavior and the candidate items respectively, inputting the extracted characteristics of the user history behavior and the extracted characteristics of the candidate items into a preset prediction model to obtain recommended indexes of the candidate items, wherein the recommended indexes of the candidate items are associated with the attention of the user, and selecting the predetermined number of candidate items as recommended results according to the sequence of the recommended indexes from large to small.
In some embodiments, inputting the extracted user historical behavioral characteristics and candidate item characteristics into a preset predictive model to obtain recommended metrics for the candidate item includes: inputting the user history behavior characteristics into a user behavior input end of a bidirectional circulating neural network, and inputting the candidate item characteristics into a candidate item input end of the bidirectional circulating neural network; inputting an output result of a user behavior output end of the bidirectional circulating neural network into an attention model so as to obtain an attention index of a user; and carrying out matching processing on the output result of the candidate item output end of the bidirectional circulating neural network by using the attention index of the user so as to obtain the recommendation index of the candidate item.
In some embodiments, the above method further comprises: in the matching process, the matching result is corrected by using auxiliary information, wherein the auxiliary information comprises at least one of user attribute information and article attribute information.
In some embodiments, the above method further comprises: and writing the recommendation result into a user history behavior log.
In some embodiments, the candidate item is associated with the user historical behavior.
In some embodiments, training data is input into the model to be trained to obtain output results, wherein the training data includes user historical behavior log information and candidate item features;
and adjusting parameters of the model to be trained according to the deviation between the output result of the model to be trained and the historical behavior log of the user so as to obtain the prediction model.
In some embodiments, the training data further includes at least one of user attribute information and item attribute information.
According to an aspect of one or more embodiments of the present disclosure, there is provided an item recommendation device, comprising: an extraction module configured to extract a predetermined number of candidate items from the item pool; the feature extraction module is configured to extract features of the user history behavior and the candidate articles respectively; a recommendation index obtaining module configured to input the extracted user historical behavior characteristics and candidate item characteristics into a preset prediction model to obtain recommendation indexes of the candidate items, wherein the recommendation indexes of the candidate items are associated with the attention of the user; and the recommendation result selection module is configured to select a predetermined number of candidate articles to be used as recommendation results according to the sequence of the recommendation indexes from large to small.
In some embodiments, the recommendation index obtaining module is configured to input the user history behavior feature into a user behavior input end of the bidirectional circulating neural network, input the candidate item feature into a candidate item input end of the bidirectional circulating neural network, input an output result of the user behavior output end of the bidirectional circulating neural network into the attention model so as to obtain an attention index of a user, and perform matching processing on the output result of the candidate item output end of the bidirectional circulating neural network by using the attention index of the user so as to obtain the recommendation index of the candidate item.
In some embodiments, the recommendation index obtaining module is further configured to modify the matching result with auxiliary information during the matching process, wherein the auxiliary information includes at least one of user attribute information and item attribute information.
In some embodiments, the apparatus further comprises: and the log updating module is configured to write the recommended result into a user history behavior log.
In some embodiments, the candidate item is associated with the user historical behavior.
In some embodiments, the apparatus further comprises: the training module is configured to input training data into a model to be trained to obtain an output result, wherein the training data comprises user historical behavior log information and candidate object characteristics, and parameters of the model to be trained are adjusted according to deviation between the output result of the model to be trained and the user historical behavior log to obtain the prediction model.
In some embodiments, the training data further includes at least one of user attribute information and item attribute information.
According to an aspect of one or more embodiments of the present disclosure, there is provided an item recommendation device, comprising: a memory configured to store instructions; a processor coupled to the memory, the processor configured to perform a method according to any of the embodiments described above based on instructions stored in the memory.
According to another aspect of one or more embodiments of the present disclosure, there is provided a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions that, when executed by a processor, implement a method as referred to in any of the embodiments above.
Other features of the present disclosure and its advantages will become apparent from the following detailed description of exemplary embodiments of the disclosure, which proceeds with reference to the accompanying drawings.
Detailed Description
The following description of the technical solutions in the embodiments of the present disclosure will be made clearly and completely with reference to the accompanying drawings in the embodiments of the present disclosure, and it is apparent that the described embodiments are only some embodiments of the present disclosure, not all embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses. All other embodiments, which can be made by one of ordinary skill in the art without inventive effort, based on the embodiments in this disclosure are intended to be within the scope of this disclosure.
The relative arrangement of the components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless it is specifically stated otherwise.
Meanwhile, it should be understood that the sizes of the respective parts shown in the drawings are not drawn in actual scale for convenience of description.
Techniques, methods, and apparatus known to one of ordinary skill in the relevant art may not be discussed in detail, but should be considered part of the specification where appropriate.
In all examples shown and discussed herein, any specific values should be construed as merely illustrative, and not a limitation. Thus, other examples of the exemplary embodiments may have different values.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further discussion thereof is necessary in subsequent figures.
FIG. 1 is an exemplary flow chart of an item recommendation method of one embodiment of the present disclosure. In some embodiments, the method steps of the present embodiments may be performed by an item recommendation device.
In step 101, a predetermined number of candidate items are extracted from a pool of items.
In some embodiments, the candidate item is associated with a user's historical behavior. For example, when a user purchases a mobile phone of a certain brand, mobile phones of other models of the brand, mobile phones of other brands, electronic products associated with the mobile phones, and the like are extracted from the object pool. In other embodiments, the user's purchasing power is evaluated based on the price at which the user purchased the cell phone, and the item associated with the purchasing power is selected from the pool of items as a candidate item. Further, in still other embodiments, candidate items may also be randomly selected from a pool of items. Thus, the user can be provided with a rich choice of items as much as possible.
In step 102, feature extraction is performed on the user's historical behavior and candidate items, respectively.
For example, the picture features may be extracted using a Deep convolutional neural network (Deep CNN), and the information features may be extracted using a Deep neural network DNN.
In step 103, the extracted historical behavior features of the user and the features of the candidate item are input into a preset prediction model to obtain a recommendation index of the candidate item, wherein the recommendation index of the candidate item is associated with the attention of the user.
By analyzing the attention of the user, the interests of the user can be clarified, and accordingly articles can be recommended to the user more accurately.
In step 104, a predetermined number of candidate items are selected as recommendation results in order of the recommendation index from the higher level to the lower level.
In some embodiments, the recommendation results are written into a user history behavior log. Because the user history behavior log participates in training of the prediction model, the recommendation result is written into the user history behavior log, so that a more accurate prediction model can be obtained.
In the item recommending method provided by the embodiment of the disclosure, the user history behavior characteristics and the candidate item characteristics are input into the preset prediction model to obtain the recommendation index of the candidate item, and the recommendation index of the candidate item is associated with the attention of the user, so that the item can be recommended to the user more accurately.
FIG. 2 is an exemplary flow chart of a prediction method of one embodiment of the present disclosure. In some embodiments, the method steps of the present embodiments may be performed by an item recommendation device.
In step 201, the user historical behavior characteristics are input into the user behavior input of the bi-directional recurrent neural network and the candidate item characteristics are input into the candidate item input of the bi-directional recurrent neural network.
Because each node of the bidirectional recurrent neural network is processed by utilizing past and future contexts, the information expressed by the historical behaviors of the user can be fully mined.
In some embodiments, the two-way recurrent neural network is a two-way long short term memory (Bi-directional Long Short Term Memory, biLSTM) network.
In step 202, the output result of the user behavior output terminal of the bidirectional recurrent neural network is input into the attention model so as to obtain the attention index of the user.
In step 203, the output result of the candidate item output end of the bidirectional recurrent neural network is matched by using the attention index of the user, so as to obtain the recommendation index of the candidate item.
FIG. 3 is a schematic diagram of a predictive model architecture according to one embodiment of the disclosure. As shown in fig. 3, the user history behavior features are input to the user behavior inputs X1, X2, …, xn of the bi-directional recurrent neural network, and the candidate item features are input to the candidate item input Xm of the bi-directional recurrent neural network. The output results of the user behavior output ends Y1, Y2, … and Yn of the bidirectional circulating neural network are input into the attention model so as to obtain the attention index Z of the user. The matching model performs matching processing on the output result of the candidate item output end Ym by using the attention index Z of the user so as to obtain the recommended index of the candidate item.
For example, the last purchase of a user was a purchase of a pen. However, by analyzing the user's historical behavior, it is found that purchasing pens is only one sporadic behavior, and the user's attention is focused on consumer electronics. In this case, the recommended value of the consumer electronics product is increased and the recommended value of the stationery is decreased when the matching process is performed.
In some embodiments, the matching result is modified with auxiliary information during the matching process, wherein the auxiliary information includes at least one of user attribute information and item attribute information.
As shown in fig. 3, the auxiliary information feature F is provided to the matching module so that the matching module corrects the matching result according to the auxiliary information when performing the matching process.
For example, if the user is a male, it is intended to recommend a black, metallic colored item to the user. If the user is female, it is intended to recommend pink, red items to the user. For another example, if the user is in the south, the air conditioner may be recommended to the user because of the high temperature and humidity in the south.
In some embodiments, training data is input into the model to be trained as the model is trained to obtain output results, wherein the training data includes user historical behavior log information and candidate item features. And adjusting parameters of the model to be trained according to the deviation between the output result of the model to be trained and the historical behavior log of the user so as to obtain a prediction model.
In some embodiments, the training data further includes at least one of user attribute information and item attribute information.
FIG. 4 is an exemplary block diagram of an item recommendation device of one embodiment of the present disclosure. As shown in fig. 4, the article recommendation device includes an extraction module 41, a feature extraction module 42, a recommendation index acquisition module 43, and a recommendation result selection module 44.
The extraction module 41 is configured to extract a predetermined number of candidate items from the pool of items.
In some embodiments, the candidate item is associated with a user's historical behavior. In other embodiments, the candidate items may be randomly selected.
The feature extraction module 42 is configured to perform feature extraction on the user's historical behavior and candidate items, respectively.
The recommendation index obtaining module 43 is configured to input the extracted user historical behavior features and candidate item features into a preset predictive model to obtain recommendation indexes of candidate items, wherein the recommendation indexes of candidate items are associated with the attention of the user.
The recommendation result selecting module 44 is configured to select a predetermined number of candidate items as recommendation results in order of the recommendation index from the higher to the lower.
In the article recommending device provided in the above embodiment of the present disclosure, the user history behavior feature and the candidate article feature are input into the preset prediction model to obtain the recommendation index of the candidate article, and the recommendation index of the candidate article is associated with the attention of the user, so that the article can be recommended to the user more accurately.
In some embodiments, the recommendation index obtaining module 43 is configured to input the user history behavior feature into the user behavior input end of the bidirectional recurrent neural network, input the candidate item feature into the candidate item input end of the bidirectional recurrent neural network, input the output result of the user behavior output end of the bidirectional recurrent neural network into the attention model so as to obtain the attention index of the user, and perform matching processing on the output result of the candidate item output end of the bidirectional recurrent neural network by using the attention index of the user so as to obtain the recommendation index of the candidate item.
In some embodiments, the recommendation index obtaining module 43 is further configured to modify the matching result with auxiliary information during the matching process, wherein the auxiliary information includes at least one of user attribute information and item attribute information.
Fig. 5 is an exemplary block diagram of an item recommendation device of another embodiment of the present disclosure. Fig. 5 differs from fig. 4 in that in the embodiment shown in fig. 5, a log update module 45 is also included.
The log update module 45 is configured to write the recommendation results into the user history behavior log. Because the user history behavior log participates in model training, the recommendation result is written into the user history behavior log, and a more accurate training result can be obtained.
In some embodiments, as shown in FIG. 5, the item recommendation device further includes a training module 46.
The training module 46 is configured to input training data into the model to be trained to obtain output results, wherein the training data includes user historical behavior log information and candidate item features, and adjust parameters of the model to be trained to obtain a prediction model according to deviations of the output results of the model to be trained from the user historical behavior log.
In some embodiments, the training data further includes at least one of user attribute information and item attribute information.
Fig. 6 is an exemplary block diagram of an item recommendation device of yet another embodiment of the present disclosure. As shown in fig. 4, the item recommendation device includes a memory 61 and a processor 62.
The memory 61 is for storing instructions and the processor 62 is coupled to the memory 61, the processor 62 being configured to perform a method as referred to in any of the embodiments of fig. 1-2 based on the instructions stored by the memory.
As shown in fig. 6, the article recommendation device further includes a communication interface 63 for information interaction with other devices. Meanwhile, the device further comprises a bus 64, and the processor 62, the communication interface 63 and the memory 61 communicate with each other through the bus 64.
The memory 61 may comprise a high-speed RAM memory or may further comprise a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 61 may also be a memory array. The memory 61 may also be partitioned and the blocks may be combined into virtual volumes according to certain rules.
Further, the processor 62 may be a central processing unit CPU, or may be an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present disclosure.
The present disclosure also relates to a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement a method as referred to in any of the embodiments of fig. 1-2.
FIG. 7 is a schematic diagram of an item recommendation framework in accordance with one embodiment of the present disclosure.
As shown in fig. 7, when recommending items online for a user, a predetermined number of candidate items are extracted from the item pool as a candidate set. And respectively extracting features of the user historical behaviors and candidate items in the candidate set, and inputting the extracted features of the user historical behaviors and the features of the candidate items into a preset prediction model to obtain recommendation indexes of the candidate items, wherein the recommendation indexes of the candidate items are associated with the attention of the user. And selecting a predetermined number of candidate articles as recommendation results according to the sequence of the recommendation indexes from large to small, and writing the recommendation results into a user history behavior log.
In the offline training process, training the model to be trained by using the user behavior log information, and adjusting parameters of the model to be trained according to the deviation between the output result of the model to be trained and the user historical behavior log so as to obtain a prediction model.
In some embodiments, the functional unit blocks described above may be implemented as general-purpose processors, programmable logic controllers (Programmable Logic Controller, abbreviated as PLCs), digital signal processors (Digital Signal Processor, abbreviated as DSPs), application specific integrated circuits (Application Specific Integrated Circuit, abbreviated as ASICs), field programmable gate arrays (Field-Programmable Gate Array, abbreviated as FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any suitable combination thereof for performing the functions described in the present disclosure.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The description of the present disclosure has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the disclosure in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments were chosen and described in order to best explain the principles of the disclosure and the practical application, and to enable others of ordinary skill in the art to understand the disclosure for various embodiments with various modifications as are suited to the particular use contemplated.