CN112258297A - Method, device and computer-readable storage medium for pushing description information of article - Google Patents
Method, device and computer-readable storage medium for pushing description information of article Download PDFInfo
- Publication number
- CN112258297A CN112258297A CN202011263515.2A CN202011263515A CN112258297A CN 112258297 A CN112258297 A CN 112258297A CN 202011263515 A CN202011263515 A CN 202011263515A CN 112258297 A CN112258297 A CN 112258297A
- Authority
- CN
- China
- Prior art keywords
- description information
- item
- article
- historical
- target user
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000003062 neural network model Methods 0.000 claims description 28
- 230000015654 memory Effects 0.000 claims description 15
- 230000006399 behavior Effects 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 10
- 238000010586 diagram Methods 0.000 description 20
- 230000006870 function Effects 0.000 description 8
- 230000009471 action Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002349 favourable effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0623—Item investigation
- G06Q30/0625—Directed, with specific intent or strategy
- G06Q30/0627—Directed, with specific intent or strategy using item specifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The disclosure provides a method and a device for pushing description information of an article and a computer-readable storage medium, and relates to the technical field of computers. In some embodiments of the present disclosure, one or more target users corresponding to the item are selected; generating candidate description information of the article according to the article; determining the selection rate of each target user for the article according to the attribute of the article, the attribute of each target user and the candidate description information of the article; then, according to the selection rate of each target user for the article, determining candidate description information with the selection rate meeting a preset condition as article description information of the article relative to the target user; and finally, respectively pushing the article description information of the article relative to each target user. Different article names are generated for different users, personalized article description information is pushed for different users, and user experience is improved.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a method and an apparatus for pushing description information of an article, and a computer-readable storage medium.
Background
With the continuous development of internet technology, the influence of e-commerce platforms on the daily life of people is deepened day by day. The user can quickly know the information of the article based on the description information of the article pushed by the E-commerce platform. In some related art, for one item, description information of the item is determined, and then the description information of the item is transmitted to all users.
Disclosure of Invention
The inventor finds that, in the related art, for an article, the description information of the article pushed to all users is the same, but the points of interest or points of interest of different users for the same article may be different, and therefore, the same description information of the article is difficult to meet the personalized needs of different users, the user experience is not good, and the possibility of selecting the article by the user is even reduced due to the difficulty of attracting the attention or interest of the user.
It is considered that, in the case where there is a difference in the description information for the same item, the feeling and selection of the item by different users are affected accordingly. In order to improve the user experience, the description information of different articles can be determined for different users, and personalized article description information can be pushed for different users, so that the description information of the articles suitable for the attention points or interest points of different users can be found, the user experience can be improved, and the article selection rate of the users can be improved.
To this end, the present disclosure provides a method capable of pushing description information of personalized items for a user.
In some embodiments of the present disclosure, one or more target users corresponding to the item are selected; generating candidate description information of the article according to the article; determining the selection rate of each target user for the article according to the attribute of the article, the attribute of each target user and the candidate description information of the article; then, according to the selection rate of each target user for the article, determining candidate description information with the selection rate meeting a preset condition as article description information of the article relative to the target user; and finally, respectively pushing the article description information of the article relative to each target user. Different article names are generated for different users, personalized article description information is pushed for different users, and user experience is improved.
According to some embodiments of the present disclosure, there is provided a method for pushing item description information, including:
selecting one or more target users corresponding to the article;
generating candidate description information of the item according to the item;
determining the selection rate of each target user for the item according to the attribute of the item, the attribute of each target user and the candidate description information of the item;
according to the selection rate of each target user for the article, determining candidate description information with the selection rate meeting the condition as article description information of the article relative to the target user;
and respectively pushing the article description information of the article relative to each target user.
In some embodiments, selecting one or more target users for which the item corresponds comprises: and according to the attribute of the historical user associated with the item, determining the user with the similarity meeting the condition with the attribute of the historical user as the target user.
In some embodiments, selecting one or more target users for which the item corresponds comprises: inputting the attribute of the article into a target user generation model, and acquiring a target user corresponding to the article output by the target user generation model; the target user generation model is obtained by training a neural network model by utilizing the attributes of the article, the attributes of the historical user associated with the article and the behavior information of the historical user on the article.
In some embodiments, the generating candidate description information for the item comprises: inputting the attribute of the article and the attribute of each target user into a generator in a generating-confrontation model, and acquiring candidate description information of the article relative to each target user, which is output by the generator.
In some embodiments, the generative-antagonistic model is trained by: inputting the attributes of the historical items in the training set and the attributes of the historical users associated with the historical items into a generator in a generation-confrontation model; acquiring one or more pieces of prediction description information of each historical item relative to each historical user output by the generator; determining a first loss of the generator according to the predicted description information of each historical item relative to each historical user and the real description information of each historical item relative to each historical user; inputting the real description information and the first real label of each historical article, and the prediction description information and the second real label of each historical article relative to each historical user into a discriminator in a generating-confrontation model; acquiring a first discrimination label aiming at real description information and a second discrimination label aiming at prediction description information which are output by the discriminator; determining a second loss of the discriminator according to the first discrimination label, the first real label, the second discrimination label and the second real label; and updating the parameters of the generator according to the first loss of the generator, updating the parameters of the discriminator according to the second loss of the discriminator until a preset condition is met, finishing training, and taking the generator and the discriminator which are finished training as the generation-confrontation model.
In some embodiments, the generating candidate description information for the item comprises: and performing word segmentation processing on the original description information of the article to obtain one or more permutation and combination of the word segmentation, and taking the permutation and combination as candidate description information of the article.
In some embodiments, said determining a selection rate of said item by said each target user comprises: inputting the attribute of the article, the attribute of each target user and the candidate description information of the article into a selection rate model, and acquiring the selection rate of each target user for the article, which is output by the selection rate model; the selection rate model is obtained by training a neural network model by utilizing the attributes of historical articles, the attributes of historical users associated with the historical articles, the real description information of the historical articles and the real selection rates of the historical users on the historical articles.
In some embodiments, the selection rate satisfying a preset condition includes: the selection rate reaches a preset threshold value, or the rank of the selection rate reaches a preset rank.
According to still other embodiments of the present disclosure, there is provided an apparatus for pushing item description information, including:
a selection module configured to select one or more target users corresponding to the item;
a generation module configured to generate candidate description information of the item according to the item;
a selection rate determining module configured to determine a selection rate of the item by each target user according to the attribute of the item, the attribute of each target user and the candidate description information of the item;
the item description information determining module is configured to determine candidate description information with a selection rate meeting a preset condition as item description information of the item relative to each target user according to the selection rate of each target user on the item;
and the pushing module is configured to push the item description information of the item relative to each target user respectively.
According to still other embodiments of the present disclosure, there is provided an apparatus for pushing item description information, including: a memory; and a processor coupled to the memory, the processor configured to execute the method of pushing item description information of any embodiment based on instructions stored in the memory.
According to still further embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the method for pushing item description information according to any of the embodiments.
Drawings
The drawings that will be used in the description of the embodiments or the related art will be briefly described below. The present disclosure can be understood more clearly from the following detailed description, which proceeds with reference to the accompanying drawings.
It is to be understood that the drawings in the following description are merely exemplary of the disclosure, and that other drawings may be derived from those drawings by one of ordinary skill in the art without undue inventive faculty.
Fig. 1 illustrates a flow diagram of a method of pushing description information for an item, in accordance with some embodiments of the present disclosure.
FIG. 2 illustrates a flow diagram of a method of obtaining a target user generated model, according to some embodiments of the present disclosure.
Fig. 3 illustrates a flow diagram of a method of obtaining a generative-antagonistic model, according to some embodiments of the present disclosure.
Fig. 4 illustrates a flow diagram of a method of obtaining a selectivity model in accordance with some embodiments of the present disclosure.
Fig. 5 illustrates a schematic diagram of an apparatus for pushing description information of an item, according to some embodiments of the present disclosure.
Fig. 6 shows a schematic view of an apparatus for pushing description information of an item according to further embodiments of the present disclosure.
Detailed Description
The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure.
Fig. 1 illustrates a flow diagram of a method of pushing description information for an item, in accordance with some embodiments of the present disclosure. The method may be performed, for example, by an apparatus that pushes descriptive information for an item.
As shown in FIG. 1, the method of this embodiment includes steps 110-150.
At step 110, one or more target users (also referred to as potential users) corresponding to the item are selected.
In some embodiments, according to the attribute of the historical user associated with the item, the user whose similarity with the attribute of the historical user meets the preset condition is determined as the target user. The history user associated with the item may be, for example, a history user who has selected (e.g., purchased, clicked, collected, or shared) the item, and according to the attributes of the history users (e.g., the portrait attributes of the history users), a user whose similarity to the attributes of the history users, such as the portrait attributes, reaches a preset threshold is selected as the target user of the item. The method for calculating the similarity may be, for example, by calculating a euclidean distance, a hamming distance, a cosine distance, and the like between the attributes of the users, and is not limited to the illustrated example. The step of determining the target user may be implemented by a hooklike recommendation model, for example.
In other embodiments, the attributes of the item are input into the target user-generated model, and the target user corresponding to the item output by the target user-generated model is obtained. The target user generation model is obtained by training the neural network model by utilizing the attributes of the articles, the attributes of the historical users associated with the articles and the behavior information of the historical users on the articles. FIG. 2 illustrates a flow diagram of a method of obtaining a target user generated model, according to some embodiments of the present disclosure. As will be described in detail later.
For example, for the item G, three target users U1, U2, and U3 corresponding thereto are selected.
The method and the device have the advantages that the corresponding target users are selected for different objects, a foundation is laid for subsequently pushing the personalized description information of the object directionally to the target users, the selection rate of the object is favorably improved, and meanwhile, the directional pushing is also favorable for improving the experience of the user.
At step 120, candidate description information for the item is generated based on the item.
In some embodiments, generating candidate description information for the item comprises: inputting the attribute of the article and the attribute of each target user into a generator in the generation-confrontation model, and acquiring the candidate description information of the article relative to each target user, which is output by the generator. Fig. 3 illustrates a flow diagram of a method of obtaining a generative-antagonistic model, according to some embodiments of the present disclosure. As will be described in detail later. And aiming at the same article, inputting different attributes of target users, wherein the output candidate description information is different, and a foundation is laid for subsequently pushing personalized article description information.
For example, the attribute of the item G and the attribute of the target user U1 are input into the generator in the generation-confrontation model, and the output result of the generator is obtained 2 times, so that candidate description information of the item G relative to the target user U1 is obtained as D1 and D2. Similarly, the attribute of the item G and the attribute of the target user U2 are input into the generator in the generation-confrontation model, and the output result of the generator is obtained 2 times, so that candidate description information of the item G relative to the target user U2 is obtained as D3 and D4. Inputting the attribute of the article G and the attribute of the target user U1 into a generator in the generation-confrontation model, and obtaining the output result of the generator 2 times to obtain candidate description information of the article G relative to the target user U1, namely D5 and D6.
In other embodiments, generating candidate description information for the item includes: and performing word segmentation processing on the original description information of the article to obtain one or more permutation and combination of the word segmentation, and taking the one or more permutation and combination as candidate description information of the article. The output candidate description information is different for different articles, and a foundation is laid for subsequently pushing personalized article description information.
For example, if the original description information of the article G is subjected to word segmentation processing to obtain 5 words, and the 5 words are arranged and combined, then the word segmentation processing can be obtained
A permutation and combinationThat is, 120 pieces of candidate description information D1 'to D120' of the article G can be obtained.
In step 130, the selection rate of each target user for the item is determined according to the attribute of the item, the attribute of each target user and the candidate description information of the item.
The selection rate indicates, for example, a ratio of the number of times of completing a selection action (e.g., clicking, placing an order, collecting or sharing, etc.) to the number of times of browsing, exposing, etc., actions within a statistical period. In this embodiment, the selection rate of the object user for the item indicates the probability that the object user selects the item.
In some embodiments, the attribute of the item, the attribute of each target user and the candidate description information of the item are input into the selection rate model, and the selection rate of each target user for the item output by the selection rate model is obtained. The selection rate model is obtained by training the neural network model by utilizing the attributes of the historical articles, the attributes of the historical users associated with the historical articles and the real description information of the historical articles. Fig. 4 illustrates a flow diagram of a method of obtaining a selectivity model in accordance with some embodiments of the present disclosure. As will be described in detail later.
By utilizing the selection rate model, aiming at the same article, if the input target users are different, the output selection rates are also different, and support is provided for pushing the description information of the personalized article for different users subsequently.
If the candidate description information of the article is generated through the generation-confrontation model, for example, the candidate description information of the article G relative to the target user U1 is obtained as D1 and D2, the attribute of the article G, the attribute of the target user U1 and the candidate description information D1 and D2 of the article are input into the selection rate model, and 2 selection rates S1 and S2 are obtained respectively. Similarly, 2 selection rates S3 and S4 of the target user U2 for the item G and 2 selection rates S5 and S6 of the target user U3 for the item G may be obtained, as shown in table 1:
TABLE 1
If the candidate description information of the article is generated by the method of arranging and combining the participles of the original description information of the article, for example, 120 candidate description information of the target user U1 for the article G can be obtained by corresponding to 120 selection rates S1 'to S120'. Similarly, 120 selection rates S121 'to S240' of the target user U2 for the item G may be obtained, and 120 selection rates S241 'to S360' of the target user U3 for the item G may be obtained.
In step 140, according to the selection rate of each target user for the item, the candidate description information with the selection rate satisfying the preset condition is determined as the description information of the item relative to the item of the target user.
Wherein, the selection rate meeting the preset conditions comprises: the selection rate reaches a preset threshold value, or the rank of the selection rate reaches a preset rank. In some embodiments, for example, an evolutionary computing method may be used to screen candidate description information with a selection rate satisfying a preset condition from a plurality of selection rates, and determine the candidate description information satisfying the preset condition as description information of an article relative to an article of a target user.
For example, if candidate description information D1, D2 of an item is generated using the method of generating a confrontation model, from the selection rates S1, S2 of the candidate description information D1, D2 of the target user U1 for the item, the candidate description information S1 with the first selection rate rank is screened out as personalized description information of the item G with respect to the target user U1.
For example, if the candidate description information D1 'to D120' of the article is generated by using the method of arranging and combining the participles of the original description information of the article, the candidate description information with the selection rate reaching the preset threshold value 0.8 is screened out from the selection rates S1 'to S120' of the candidate description information D1 'to D120' of the article by the target user U2, for example, the candidate description information with the selection rate reaching the preset threshold value 0.8 includes S1 'to S10', and the candidate description information S1 'to S10' is used as the personalized description information of the article G relative to the target user U2.
The candidate description information of the corresponding article with the selection rate meeting the preset condition is determined as the final description information of the article, so that the effectiveness and accuracy of subsequent pushing can be improved, the selection rate of the target user on the article is improved, and the user experience is improved.
In step 150, the description information of the item relative to the item of each target user is pushed to each target user.
For example, pushing personalized description information S1 of the item G with respect to the item of the target user U1 to the target user 1; and pushing personalized description information S1 '-S10' of the item G relative to the item of the target user U2 to the target user 2.
In the embodiment, the description information of different articles is determined for different users, and the personalized article description information is pushed for different users, so that the description information of the articles suitable for the attention points or interest points of different users can be found, the user experience is improved, and the article selection rate of the users is improved.
FIG. 2 illustrates a flow diagram of a method of obtaining a target user generated model, according to some embodiments of the present disclosure.
The target user generation model is obtained by training the neural network model by utilizing the attributes of the articles, the attributes of the historical users associated with the articles and the behavior information of the historical users on the articles.
As shown in FIG. 2, the method of this embodiment includes steps 210 and 230.
In step 210, the attribute of the item, the attribute of the historical user associated with the item, and the behavior information of the historical user on the item are input into the neural network model, and the predicted behavior information output by the neural network model is obtained.
The neural network model may be, for example, one of a Long Short-Term Memory network (LSTM) model, a deep neural network model, a convolutional neural network model, or a cyclic neural network model.
In step 220, the loss of the neural network model is determined according to the input historical user behavior information on the article and the predicted behavior information.
The behavior information of the historical user on the article is equivalent to the real label of the historical user on the article. Accordingly, the obtained predicted behavior information output by the generator corresponds to a predicted label of the item for the historical user. For example, the user's behavior information (e.g., labeled as UL) for the item based on the entered historyiI ≦ N1 ≦ N, N being the number of pieces of training data) and predicted behavior information (e.g., labeled UP)i1 ≦ i ≦ N), the Loss for determining the neural network model may be expressed, for example, as:
in step 230, parameters of the neural network model are updated according to the determined loss until a termination condition is met, training is completed, and the trained neural network model is used as a target user generation model.
The termination condition may be, for example, that the value of the loss function reaches a threshold, that the loss function converges, or that the number of iterations reaches a number.
Through the embodiment, the target user model can be obtained, the object is input into the target user model, and one or more target users of the object can be obtained by using the model. Therefore, by using the target user model, a target user (namely a potential user) of the object can be conveniently determined, and the description information of the personalized object is subsequently pushed to the target user, so that the user experience can be improved.
Fig. 3 illustrates a flow diagram of a method of obtaining a generative-antagonistic model, according to some embodiments of the present disclosure.
As shown in fig. 3, the method of this embodiment includes step 310-370. The generative-antagonistic model was trained by the following steps.
In step 310, the attributes of the historical items in the training set and the attributes of the historical users associated with the historical items are input into a generator in the generative-confrontational model, and one or more pieces of prediction description information of each historical item relative to each historical user output by the generator are obtained.
At step 320, a first loss of the generator is determined based on the predicted descriptive information of each historical item relative to each historical user and the actual descriptive information of each historical item relative to each historical user.
Based on predicted descriptive information (e.g. labeled as MP) of each historical item relative to each historical useriI ≦ N1, N being the number of training data for the historical user associated with the historical item) and the actual descriptive information (e.g., labeled ML) of each historical item relative to each historical useri1 ≦ i ≦ N), the first loss L2 of the decision generator may be expressed, for example, as:
in step 330, the real description information and the first real label of each historical item, and the predicted description information and the second real label of each historical item relative to each historical user are input into a discriminator in the generation-confrontation model, and a first discrimination label for the real description information and a second discrimination label for the predicted description information output by the discriminator are obtained.
The first real label of the real description information of the historical item is 1, for example, if the label is 1, the description information is real description information, and the second real label of the historical item relative to the prediction description information of the historical user is 0, for example, if the label is 0, the description information is non-real description information.
At step 340, a second loss of the discriminator is determined based on the first discrimination label and the first real label, the second discrimination label and the second real label.
According to a discrimination tag (e.g. labelled DP)i1 ≦ i ≦ N) and a true tag (e.g., labeled DLi1 ≦ i ≦ N, N being the number of pieces of training data of the historical user associated with the historical item), the second loss L3 of the determination discriminator may be expressed as:
wherein, when DPiDL is the first discrimination labeliIs a first real label; accordingly, when DP isiDL is the second discrimination tagiIs the second real label.
In step 350, the parameters of the generator are updated according to the first loss of the generator, the parameters of the discriminator are updated according to the second loss of the discriminator until the preset condition is met, the training is completed, and the generator and the discriminator which are trained are used as a generation-confrontation model.
Wherein, the closer the discrimination label of the discriminator to the input prediction description information is to 1, the closer the prediction description information is to the real description information, that is, the better the effect of the description generator is.
The preset condition may be, for example, that the first loss of the generator satisfies the preset condition (e.g., the first loss converges, or the number of iterations reaches the preset number), and/or that the second loss of the discriminator satisfies the preset condition (e.g., the second loss converges, or the number of iterations reaches the preset number).
The embodiment can obtain the generation-confrontation model, and the trained generation-confrontation model can be used for generating the candidate description information of the article, thereby laying a foundation for subsequently pushing the description information of the personalized article to different users.
Fig. 4 illustrates a flow diagram of a method of obtaining a selectivity model in accordance with some embodiments of the present disclosure.
The selection rate model is obtained by training the neural network model by utilizing the attributes of the historical articles, the attributes of the historical users associated with the historical articles, the real description information of the historical articles and the real selection rates of the historical users on the historical articles. The neural network model may be, for example, one of a deep neural network model, a convolutional neural network model, or a recurrent neural network model.
As shown in FIG. 4, the method of this embodiment includes steps 410-440.
In step 410, the attribute of the historical item, the attribute of the historical user associated with the historical item and the real description information of the historical item are input into the neural network model, and the predicted selection rate of the historical user on the historical item output by the neural network model is obtained.
In step 420, the actual selection rate of the historical item by the historical user is determined according to the behavior information of the historical item by the historical user associated with the historical item.
For example, the number of history users who select (e.g., click, collect, place, share, etc.) the history item from among the history users associated with the history item is counted, the number of history users who expose or browse the history item from among the history users associated with the history item is counted, and the ratio of the number of history users who select the history item to the number of history users who expose or browse the history item is calculated as the true selection rate of the history users for the history item.
At step 430, a loss of the neural network model is determined based on the historical user's true selection rate and the predicted selection rate of the historical item.
Based on the actual selection rate of historical items by the historical user (e.g. labeled as SP)i1 ≦ i ≦ N, N being the number of training data for the historical user associated with the historical item) and the predicted selection rate (e.g., labeled SLi1 ≦ i ≦ N), the loss L4 that determines the neural network model may be expressed, for example, as:
in step 440, parameters of the neural network model are updated according to the determined loss until a termination condition is satisfied, training is completed, and the trained neural network model is used as a selectivity model.
The termination condition may be, for example, that the value of the loss function reaches a preset threshold, that the loss function converges, or that the number of iterations reaches a preset number.
The selection rate model can be obtained through the embodiment, the selection rate of the target user for the object can be obtained through the trained selection rate model, the selection probability of different users for the same object has certain correlation with the description information of the object, and the personalized description information of the object can be pushed for different users by combining the selection rate.
Fig. 5 shows a schematic view of an apparatus for pushing description information of an item according to further embodiments of the present disclosure.
As shown in fig. 5, the apparatus 500 for pushing description information of an article of this embodiment includes: a selection module 510, a generation module 520, a selection rate determination module 530, an item description information determination module 540, and a push module 550.
A selection module 510 configured to select one or more target users to which the item corresponds.
In some embodiments, the selecting module 510 is configured to determine, as the target user, a user whose similarity to the attribute of the historical user satisfies a preset condition according to the attribute of the historical user associated with the item. The historical user associated with the item may be, for example, a historical user who purchased the item, and a user whose similarity to the attribute or the image attribute of the historical user reaches a preset threshold is selected as a target user of the item according to the attribute (for example, the image attribute of the historical user) of the historical user. The method for calculating the similarity may be, for example, by calculating a euclidean distance, a hamming distance, a cosine distance, and the like between the attributes of the users, and is not limited to the illustrated example.
In other embodiments, the selection module 510 is configured to input the attribute of the item into the target user-generated model, and obtain the target user corresponding to the item output by the target user-generated model. The target user generation model is obtained by training the neural network model by utilizing the attributes of the articles, the attributes of the historical users associated with the articles and the behavior information of the historical users on the articles.
A generating module 520 configured to generate candidate description information of the item according to the item.
In some embodiments, the generating module 520 is configured to input the attribute of the item and the attribute of each target user into a generator in the generation-confrontation model, and obtain candidate description information of the item relative to each target user output by the generator.
In other embodiments, the generating module 520 is configured to perform word segmentation on the original description information of the article to obtain one or more permutation and combination of the word segmentation, and use the one or more permutation and combination as candidate description information of the article.
And a selection rate determining module 530 configured to determine a selection rate of each target user for the item according to the attribute of the item, the attribute of each target user and the candidate description information of the item.
In some embodiments, the selection rate determining module 530 is configured to input the attribute of the item, the attribute of each target user, and the candidate description information of the item into the selection rate model, and obtain the selection rate of the item by each target user output by the selection rate model. The selection rate model is obtained by training the neural network model by utilizing the attributes of the historical articles, the attributes of the historical users associated with the historical articles and the real description information of the historical articles. The selection rate indicates, for example, a ratio of the number of times of completing a selection action (e.g., clicking, placing an order, collecting or sharing, etc.) to the number of times of browsing, exposing, etc., actions within a statistical period. In this embodiment, the selection rate of the object user for the item indicates the probability that the object user selects the item.
And an item description information determining module 540 configured to determine, according to the selection rate of each target user for the item, candidate description information of which the selection rate satisfies a preset condition as description information of the item relative to the item of the target user.
And the pushing module 550 is configured to push the description information of the item relative to the item of each target user to each target user.
The device for pushing the description information of the article in the embodiment can be used for pushing personalized article description information for different users, is favorable for improving the selection rate of the user on the article, and can improve the user experience.
Fig. 6 shows a schematic view of an apparatus for pushing description information of an item according to further embodiments of the present disclosure.
As shown in fig. 6, the apparatus 600 for pushing description information of an article of this embodiment includes: a memory 610 and a processor 620 coupled to the memory 610, the processor 620 configured to execute a method of pushing description information of an item in any of the embodiments of the present disclosure based on instructions stored in the memory 610.
The apparatus for pushing description information of an item 600 may further include an input-output interface 630, a network interface 640, a storage interface 650, and the like. These interfaces 630, 640, 650 and the connections between the memory 610 and the processor 620 may be, for example, via a bus 660. The input/output interface 630 provides a connection interface for input/output devices such as a display, a mouse, a keyboard, and a touch screen. The network interface 640 provides a connection interface for various networking devices. The storage interface 650 provides a connection interface for external storage devices such as an SD card and a usb disk.
As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-non-transitory readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer program code embodied therein.
The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only exemplary of the present disclosure and is not intended to limit the present disclosure, which is to be construed in any way as imposing limitations thereon, such as the appended claims, and all changes and equivalents that fall within the true spirit and scope of the present disclosure.
Claims (11)
1. A method for pushing item description information, comprising:
selecting one or more target users corresponding to the article;
generating candidate description information of the item according to the item;
determining the selection rate of each target user for the item according to the attribute of the item, the attribute of each target user and the candidate description information of the item;
according to the selection rate of each target user for the article, determining candidate description information with the selection rate meeting the condition as article description information of the article relative to the target user;
and respectively pushing the article description information of the article relative to each target user.
2. The method for pushing item description information according to claim 1, wherein selecting one or more target users corresponding to the item comprises:
and according to the attribute of the historical user associated with the item, determining the user with the similarity meeting the condition with the attribute of the historical user as the target user.
3. The method for pushing item description information according to claim 1, wherein selecting one or more target users corresponding to the item comprises:
inputting the attribute of the article into a target user generation model, and acquiring a target user corresponding to the article output by the target user generation model;
the target user generation model is obtained by training a neural network model by utilizing the attributes of the article, the attributes of the historical user associated with the article and the behavior information of the historical user on the article.
4. The method for pushing item description information according to claim 1, wherein said generating candidate description information of said item comprises:
inputting the attribute of the article and the attribute of each target user into a generator in a generating-confrontation model, and acquiring candidate description information of the article relative to each target user, which is output by the generator.
5. The method for pushing item description information as claimed in claim 4, wherein said generation-confrontation model is trained by the steps of:
inputting the attributes of the historical items in the training set and the attributes of the historical users associated with the historical items into a generator in a generation-confrontation model;
acquiring one or more pieces of prediction description information of each historical item relative to each historical user output by the generator;
determining a first loss of the generator according to the predicted description information of each historical item relative to each historical user and the real description information of each historical item relative to each historical user;
inputting the real description information and the first real label of each historical article, and the prediction description information and the second real label of each historical article relative to each historical user into a discriminator in a generating-confrontation model;
acquiring a first discrimination label aiming at real description information and a second discrimination label aiming at prediction description information which are output by the discriminator;
determining a second loss of the discriminator according to the first discrimination label, the first real label, the second discrimination label and the second real label;
and updating the parameters of the generator according to the first loss of the generator, updating the parameters of the discriminator according to the second loss of the discriminator until a preset condition is met, finishing training, and taking the generator and the discriminator which are finished training as the generation-confrontation model.
6. The method for pushing item description information according to claim 1, wherein said generating candidate description information of said item comprises:
and performing word segmentation processing on the original description information of the article to obtain one or more permutation and combination of the word segmentation, and taking the permutation and combination as candidate description information of the article.
7. The method for pushing item description information as claimed in claim 1, wherein said determining a selection rate of said item by said each target user comprises:
inputting the attribute of the article, the attribute of each target user and the candidate description information of the article into a selection rate model, and acquiring the selection rate of each target user for the article, which is output by the selection rate model;
the selection rate model is obtained by training a neural network model by utilizing the attributes of historical articles, the attributes of historical users associated with the historical articles, the real description information of the historical articles and the real selection rates of the historical users on the historical articles.
8. The method for pushing item description information according to claim 1, wherein the selection rate satisfying a preset condition comprises: the selection rate reaches a preset threshold value, or the rank of the selection rate reaches a preset rank.
9. An apparatus for pushing item description information, comprising:
a selection module configured to select one or more target users corresponding to the item;
a generation module configured to generate candidate description information of the item according to the item;
a selection rate determining module configured to determine a selection rate of the item by each target user according to the attribute of the item, the attribute of each target user and the candidate description information of the item;
the item description information determining module is configured to determine candidate description information with a selection rate meeting a preset condition as item description information of the item relative to each target user according to the selection rate of each target user on the item;
and the pushing module is configured to push the item description information of the item relative to each target user respectively.
10. An apparatus for pushing item description information, comprising:
a memory; and
a processor coupled to the memory, the processor configured to perform the method of pushing item description information of any of claims 1-8 based on instructions stored in the memory.
11. A non-transitory computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the method of pushing item description information according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011263515.2A CN112258297A (en) | 2020-11-12 | 2020-11-12 | Method, device and computer-readable storage medium for pushing description information of article |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011263515.2A CN112258297A (en) | 2020-11-12 | 2020-11-12 | Method, device and computer-readable storage medium for pushing description information of article |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112258297A true CN112258297A (en) | 2021-01-22 |
Family
ID=74265823
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011263515.2A Pending CN112258297A (en) | 2020-11-12 | 2020-11-12 | Method, device and computer-readable storage medium for pushing description information of article |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112258297A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108875820A (en) * | 2018-06-08 | 2018-11-23 | Oppo广东移动通信有限公司 | Information processing method and device, electronic equipment, computer readable storage medium |
| CN111782928A (en) * | 2019-05-20 | 2020-10-16 | 北京沃东天骏信息技术有限公司 | Information pushing method and device and computer readable storage medium |
| CN111881343A (en) * | 2020-07-07 | 2020-11-03 | Oppo广东移动通信有限公司 | Information pushing method and device, electronic equipment and computer readable storage medium |
-
2020
- 2020-11-12 CN CN202011263515.2A patent/CN112258297A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108875820A (en) * | 2018-06-08 | 2018-11-23 | Oppo广东移动通信有限公司 | Information processing method and device, electronic equipment, computer readable storage medium |
| CN111782928A (en) * | 2019-05-20 | 2020-10-16 | 北京沃东天骏信息技术有限公司 | Information pushing method and device and computer readable storage medium |
| CN111881343A (en) * | 2020-07-07 | 2020-11-03 | Oppo广东移动通信有限公司 | Information pushing method and device, electronic equipment and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210287048A1 (en) | System and method for efficient generation of machine-learning models | |
| CN107463704B (en) | Search method and device based on artificial intelligence | |
| US11403532B2 (en) | Method and system for finding a solution to a provided problem by selecting a winner in evolutionary optimization of a genetic algorithm | |
| CN109190049B (en) | Keyword recommendation method, system, electronic device and computer readable medium | |
| CN106484777B (en) | Multimedia data processing method and device | |
| CN111339427B (en) | Book information recommendation method, device and system and storage medium | |
| EP3759621A1 (en) | Content editing using ai-based content modeling | |
| CN109241243B (en) | Candidate document sorting method and device | |
| US9122710B1 (en) | Discovery of new business openings using web content analysis | |
| WO2019061664A1 (en) | Electronic device, user's internet surfing data-based product recommendation method, and storage medium | |
| CN111461345A (en) | Deep learning model training method and device | |
| CN116010684A (en) | Article recommendation method, device and storage medium | |
| CN103534696A (en) | Exploiting query click logs for domain detection in spoken language understanding | |
| CN111563198B (en) | Material recall method, device, equipment and storage medium | |
| CN114036398A (en) | Content recommendation and ranking model training method, device, equipment and storage medium | |
| CN112380104A (en) | User attribute identification method and device, electronic equipment and storage medium | |
| CN110968802B (en) | Analysis method and analysis device for user characteristics and readable storage medium | |
| CN113569118B (en) | Self-media pushing method, device, computer equipment and storage medium | |
| CN111667018B (en) | Object clustering method and device, computer readable medium and electronic equipment | |
| CN108595395B (en) | Nickname generation method, device and equipment | |
| CN112258297A (en) | Method, device and computer-readable storage medium for pushing description information of article | |
| CN115017413A (en) | Recommendation method and device, computing equipment and computer storage medium | |
| JP5555238B2 (en) | Information processing apparatus and program for Bayesian network structure learning | |
| CN113886697A (en) | Clustering algorithm based activity recommendation method, device, equipment and storage medium | |
| CN114398877A (en) | Theme extraction method and device based on artificial intelligence, electronic equipment and 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 |