CN111931055B

CN111931055B - Object recommendation method, object recommendation device and electronic equipment

Info

Publication number: CN111931055B
Application number: CN202010816884.3A
Authority: CN
Inventors: 权爱荣; 马晓楠; 王雅楠
Original assignee: Industrial and Commercial Bank of China Ltd ICBC; ICBC Technology Co Ltd
Current assignee: Industrial and Commercial Bank of China Ltd ICBC; ICBC Technology Co Ltd
Priority date: 2020-08-14
Filing date: 2020-08-14
Publication date: 2023-09-12
Anticipated expiration: 2040-08-14
Also published as: CN111931055A

Abstract

The disclosure provides an object recommendation method, an object recommendation device and electronic equipment, which can be used in the technical field of artificial intelligence, and the method comprises the following steps: respectively determining at least one first recommended object information through M recall channels to obtain a first recommended object information set, wherein M is a positive integer greater than or equal to 2; determining a recommendation score for each first recommendation object information by repeating the following operations: for each piece of first recommended object information, determining recommended object association features, and processing the recommended object association features by using the first object recommendation model to determine recommendation scores of the first recommended object information; and determining a second set of recommended object information based on the recommendation score of each of the first recommended object information to output the second set of recommended object information.

Description

Object recommendation method, object recommendation device and electronic equipment

Technical Field

The present disclosure relates to the field of artificial intelligence, and more particularly, to an object recommendation method, an object recommendation apparatus, and an electronic device and a computer-readable storage medium.

Background

With the rapid development of artificial intelligence technology, intelligent recommendation plays an increasingly important role. The intelligent recommendation method comprises the main tasks of obtaining preference characteristics of a user on an object by analyzing user information, object information or other auxiliary information, and recommending the object for the user according to the preference characteristics.

In the process of implementing the disclosed concept, the inventor finds that at least the following problems exist in the related art: the recommendation algorithm in the related art comprises a collaborative filtering algorithm, and for government scenes, the collaborative filtering algorithm has the problems of cold start and sparse user behavior data, so that the accuracy of a recommendation result is low.

Disclosure of Invention

In view of the above, the present disclosure provides an object recommendation method, an object recommendation device and an electronic device for solving the problems of cold start and sparse user behavior data in government scenes by combining multiple recall and recommendation ordering.

One aspect of the present disclosure provides an object recommendation method. The method comprises the following steps: respectively determining at least one first recommended object information through M recall channels to obtain a first recommended object information set, wherein M is a positive integer greater than or equal to 2; determining a recommendation score for each first recommendation object information by repeating the following operations: for each piece of first recommended object information, determining a recommended object association feature, and processing the recommended object association feature by using the first object recommendation model to determine a recommendation score of the first recommended object information; a second set of recommended object information is determined based on the recommendation score of each of the first recommended object information to output the second set of recommended object information.

Another aspect of the present disclosure provides an object recommendation apparatus. The device comprises: the system comprises a recall module, a recommendation score determining module and a recommendation object determining module. The recall module is used for respectively determining at least one piece of first recommended object information through M recall channels to obtain a first recommended object information set, wherein M is a positive integer greater than or equal to 2; the recommendation score determining module is used for determining a recommendation score of each piece of first recommendation object information by repeating the following operations: for each piece of first recommended object information, determining a recommended object association feature, and processing the recommended object association feature by using the first object recommendation model to determine a recommendation score of the first recommended object information; the recommended object determining module is used for determining a second recommended object information set based on the recommended score of each piece of first recommended object information so as to output the second recommended object information set.

Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more instructions, wherein the one or more instructions, when executed by the one or more processors, cause the one or more processors to implement the object recommendation method as described above.

Another aspect of the present disclosure provides a computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to implement an object recommendation method as above.

Another aspect of the present disclosure provides a computer program comprising computer executable instructions which, when executed, are for implementing an object recommendation method as described above.

According to the embodiment of the disclosure, aiming at the defect of insufficient object recommendation precision in a cold start and data sparse scene, at first, at least one piece of first recommended object information is respectively determined through M recall channels, the range of the recalled object is determined, the problem that user history data is lacking in the cold start and data sparse scene is solved, then all objects in the recalled object range are ordered through an object recommendation model to determine recommended objects, the recall rate of recommendation results in the cold start and data sparse scene is effectively improved, and the recommendation effect is improved.

Drawings

The above and other objects, features and advantages of the present disclosure will become more apparent from the following description of embodiments thereof with reference to the accompanying drawings in which:

Fig. 1 schematically illustrates an application scenario of an object recommendation method, an object recommendation apparatus, and an electronic device according to an embodiment of the present disclosure;

FIG. 2 schematically illustrates a system architecture suitable for use in an object recommendation method, an object recommendation apparatus, and an electronic device according to an embodiment of the present disclosure;

FIG. 3 schematically illustrates a flow chart of an object recommendation method according to an embodiment of the present disclosure;

FIG. 4 schematically illustrates a schematic view of a recall channel according to an embodiment of the present disclosure;

FIG. 5 schematically illustrates a structural diagram of first candidate recommended object information according to an embodiment of the present disclosure;

FIG. 6 schematically illustrates a flow chart of determining a second set of recommended object information according to an embodiment of the disclosure;

FIG. 7 schematically illustrates a schematic diagram of a recommended object association feature according to an embodiment of the disclosure;

FIG. 8 schematically illustrates a technical framework diagram of an object recommendation method according to an embodiment of the present disclosure;

FIG. 9 schematically illustrates a block diagram of an object recommendation apparatus according to an embodiment of the present disclosure; and

fig. 10 schematically illustrates a block diagram of an electronic device adapted to perform an object recommendation method according to an embodiment of the present disclosure.

Detailed Description

Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that the description is only exemplary and is not intended to limit the scope of the present disclosure. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. It may be evident, however, that one or more embodiments may be practiced without these specific details. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the concepts of the present disclosure.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The terms "comprises," "comprising," and/or the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.

All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the terms used herein should be construed to have meanings consistent with the context of the present specification and should not be construed in an idealized or overly formal manner.

Where expressions like at least one of "A, B and C, etc. are used, the expressions should generally be interpreted in accordance with the meaning as commonly understood by those skilled in the art (e.g.," a system having at least one of A, B and C "shall include, but not be limited to, a system having a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.). Where a formulation similar to at least one of "A, B or C, etc." is used, in general such a formulation should be interpreted in accordance with the ordinary understanding of one skilled in the art (e.g. "a system with at least one of A, B or C" would include but not be limited to systems with a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.). The terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more of the described features.

In order to facilitate understanding of the technical scheme of the present disclosure, first, intelligent recommendation will be described. The main task of intelligent recommendation is to obtain the preference characteristics of the object to be recommended for the user by analyzing the user information, the object information to be recommended (such as commodity information, function information, consultation information and the like) or other auxiliary information, and accordingly, to recommend the object for the user.

In the related art, the recommendation algorithm mainly includes a collaborative filtering algorithm and a content-based recommendation algorithm. Collaborative filtering is to find some similarity through the behavior of the population, and make recommendations to the user through the similarity. Collaborative filtering algorithms fall into two categories, one is to make object recommendations to a given user based on the preferences of similar users, i.e., collaborative filtering based on the user. One is to make recommendations for a given user based on the similarity of the objects, i.e., collaborative filtering recommendations based on the objects. Content-based recommendations are made by finding items of other similar properties through items that the user has selected.

The collaborative filtering algorithm has two disadvantages, namely, the problem of cold start is solved, and if the user behavior is small, such as a new online product or a product with a small user scale, the recommendation can not be made for the user. Secondly, the sparsity problem is that if the user only acts on few objects in a scene with a large number of objects, the user behavior matrix is very sparse, and the final recommendation result is not accurate enough. Government scenario users typically only transact individual items and users may not log in the product for long periods of time, so the user behavior data is very sparse.

An efficient feature extraction is required for content-based recommendation algorithms, and features of extracted items (items) can be compared for content such as text personalized reading, but it is difficult to accurately characterize items for government matters.

The embodiment of the disclosure provides an object recommendation method, an object recommendation device and electronic equipment. The method includes an object recall process and an object ordering process. In the object recall process, at least one piece of first recommended object information is respectively determined through M recall channels to obtain a first recommended object information set, wherein M is a positive integer greater than or equal to 2. Entering an object ordering process after completing the object recall process, first, repeating the following operations to determine a recommendation score for each first recommended object information: for each first recommended object information, determining a recommended object association feature, processing the recommended object association feature with the first object recommendation model to determine a recommendation score for the first recommended object information, and then determining a second recommended object information set based on the recommendation score for each first recommended object information to output the second recommended object information set. And in the cold start and data sparse scene of the to-be-handled government matters and the like, the recall rate of the recommended results of the government matters and the like is effectively improved through the combined use of the multipath recall and the recommended sequencing, and the recommended effect is improved.

Fig. 1 schematically illustrates an application scenario of an object recommendation method, an object recommendation apparatus, and an electronic device according to an embodiment of the present disclosure. It should be noted that, the object recommendation method, the object recommendation device and the electronic device according to the embodiments of the present disclosure may be used in the field of artificial intelligence technology, and may also be used in various fields other than the field of artificial intelligence technology, such as the field of big data technology. The application fields of the object recommendation method, the object recommendation device and the electronic equipment in the embodiment of the disclosure are not limited.

As shown in fig. 1, taking a government Application (APP) in a mobile phone as an example, the application may include a plurality of display areas, where one area may be set to display a personalized recommendation object. The uppermost display area in fig. 1 is a personalized recommended object display area for the current login user, and the rest areas can include display areas of preset functions of the application, such as a flood prevention special area, a policy file library and the like. Wherein the personalized recommended object display area is limited in area, and in order to facilitate the user to quickly find a desired object, only a specified number of object sub-display areas are generally provided in the personalized recommended object display area, such as 3, 5, 8, 10, 11, 15, 20, etc. Wherein the personalized recommended object may be determined based on user attributes, historical operations, object attributes, and the like.

However, in the case of government matters, there are much less user operation data than in the case of online purchase of commodities and the like, and there are a cold start problem and a data sparseness problem. Such as government scenarios where users typically only transact individual items and may not log in the product for long periods of time, the user behavior data is very sparse. In addition, content-based recommendation algorithms require efficient feature extraction, but it is difficult to accurately characterize items for government matters. According to the object recommending method, the object recommending device and the electronic equipment, the problem that user history data is lacking in a cold start and data sparse scene is solved through the multi-way recall and the object recommending ordering method, the recall rate of a recommending result in the cold start and data sparse scene is effectively improved, and the recommending effect is further improved.

Fig. 2 schematically illustrates a system architecture suitable for use in an object recommendation method, an object recommendation apparatus, and an electronic device according to an embodiment of the present disclosure. It should be noted that fig. 2 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios.

As shown in fig. 2, a system architecture 200 according to an embodiment of the present disclosure may include terminal devices 201, 202, 203, a network 204, a server 205. The terminal devices 201, 202, 203 and the server 205 may be connected through a network 204, where the network 204 may include various connection types, such as a wired, wireless communication link, or a fiber optic cable, etc.

The terminal devices 201, 202, 203 may be provided with a display screen and/or may be provided with various programs such as client applications, software, etc., including but not limited to smartphones, tablet computers, laptop portable computers, mainframe and desktop computers, etc. The terminal device 201, 202, 203 may provide multiple functional portals to the user through a client application to meet multiple needs of the user.

According to an embodiment of the present disclosure, the terminal device 201, 202, 203 may further have a processing function to process an object or the like to be recommended to the user, based on the user attribute information, the user history operation, the object attribute information, or the like. To preferentially present interactive components (e.g., function icons, etc.) corresponding to these recommended objects in the display interface of the client application. As shown with reference to fig. 2, a plurality of government matters, etc. may be included.

The network 204 is used as a medium to provide communication links between terminal devices and servers. The network may include various connection types, such as wired and/or wireless communication links, and the like.

The server 205 may be a server providing various services. For example, the server 205 may interact with the terminal devices 201, 202, 203 through the network 204 to acquire user attribute information, user history operation information, and the like. For example, the terminal apparatuses 201, 202, 203 may interact with the server 205 through the network 204 to acquire object attribute information and the like. The terminal device 201, 202, 203 or the server 205 may process user attribute information, user history operation information, object attribute information, or the like to obtain an object to be recommended to the user.

It should be noted that, the object recommendation method provided by the embodiments of the present disclosure may be performed by the terminal device 201, 202, 203 or the server 205. Accordingly, the object recommendation apparatus provided by the embodiments of the present disclosure may be provided in the terminal device 201, 202, 203 or the server 205. It should be understood that the types and numbers of terminal devices 201, 202, 203, network 204, and server 205 in fig. 2 are merely illustrative.

Fig. 3 schematically illustrates a flowchart of an object recommendation method according to an embodiment of the present disclosure. It should be noted that, the object recommendation method may be performed offline or online. An offline object recommendation method will be described. The following collaborative filtering refers to: information of interest to the user is recommended using interest in the interest group. The following recalls are referred to as: as many correct results as possible are triggered from the full set of information.

As shown in fig. 3, the object recommendation method performed by the server side may include operations S301 to S305.

In operation S301, at least one first recommended object information is determined through M recall channels, respectively, to obtain a first recommended object information set, where M is a positive integer greater than or equal to 2.

The number M of recall channels may be empirical or scene dependent. The elements which can possibly have relation with the object to be recommended can be respectively abstracted into a recall channel. For example, if the user search operation may be for an object, then it may be determined that a search term recall channel may be used. For example, if a user's question at an intelligent customer service may be for an object, it may be determined that a question-and-answer system event recall channel may be used. For example, if the probability that the user wishes to quickly find a trending event is high, then a trending event recall channel may be determined. In addition, object recall may also be performed by integrating multiple elements. For example, recall channels, content-based features, etc. may be filtered through collaborative filtering.

In one embodiment, to better address the issues of cold start, sparse data, etc. while improving the comprehensiveness and accuracy of recall, the recall channel may include a collaborative filtering recall channel, a user behavior weight statistics recall channel, and at least one of the following: search item recall channel, consultation item recall channel, question and answer system item recall channel, or topical item recall channel. The collaborative filtering recall channel and the user behavior weight statistics recall channel can effectively improve the object recall rate and accuracy in non-cold start and non-data sparse scenes. The search item recall channel, the consultation item recall channel and the question and answer system item recall channel can be used for recalling strongly related items by a user operation, for example, if the user searches coronaviruses, epidemic prevention and control can be predicted to be government affair items which the user hopes to find quickly. As another example, a user asking in a smart client which recruits all recently, a employment may be predicted to be a government event that the user wishes to quickly find. Although the user does not generate a large amount of user operation data or history data, the problems of cold start and data sparseness can be solved based on government matters that the user may need to use, such as user search and questioning. In addition, in order to further improve the comprehensiveness of the recall of the objects, when the number of the objects recalled through the previous recall channels is small, the recall channels can be supplemented through the results of hot events, the hot events are usually generated due to the fact that more users pay attention together, and the demands of more users can be generally reflected, so that the supplement of the hot events is helpful for improving the recall comprehensiveness in the cold start and data sparse scene.

In one embodiment, determining the respective at least one first recommended object information through the M recall channels, respectively, may include the following operations.

For example, for a collaborative filtering recall channel, at least one first recommended object information is determined based on a collaborative filtering algorithm.

For example, for a user behavior weight statistics recall channel, the following operations are performed for each user separately: and counting the behaviors of the user aiming at each object information respectively based on the behavior weights to obtain the statistical scores of the user aiming at each object information respectively, and determining at least one first recommended object information based on the statistical scores.

For example, for at least one channel of the consultation item recall channel, the question and answer system item recall channel, or the popular item recall channel, at least one first recommendation object information of each channel is determined by keyword matching of the word segmentation result of the input information.

For example, for the hot event recall channel, the first Q of the hot events counted in the specified period are taken as at least one first recommendation object information, Q is a positive integer greater than or equal to zero.

The following description will take 6 recall channels as an example.

Fig. 4 schematically illustrates a schematic view of a recall channel according to an embodiment of the present disclosure.

As shown in fig. 4, for a transaction-based collaborative filtering algorithm (e.g., user-based collaborative filtering). And analyzing other users similar to the user behaviors according to the user behaviors, and recommending matters preferred by the other users.

For rule algorithms based on user behavior weight statistics. And (3) giving different weights to the transaction behaviors of different matters according to the service, such as clicking behavior weight 1, searching behavior weight 2, submitting behavior weight 3, collecting behavior weight 3 and the like. Separately counting different behaviors of each user, grading and sorting the matters, and selecting the matters with the higher score to the lower score of the transacted matters of the user for recommendation according to the statistical data; for example, the user clicks once to record a weight of 1; a record weight of 2 is submitted, and finally the item score=number of times is counted for the user, for example, 1×1+1×2=3. This makes it possible to obtain a score for each item for the current user, so as to determine which items to recall based on the score. The use frequency and habit of the user are counted from the historical behavior of the user handling government affairs, the frequently handled matters of the user are recommended, and the recall rate of the recommendation based on the user behavior counting rule algorithm is about 21% in one month on line from the on-line counting, which is higher than that of the single use of the collaborative filtering algorithm.

For hot event based recall. According to project data statistics, the first N (TopN) of the popular items counted on the day is used as recommended items of all users, the recommended recall rate is about 18%, and the current popular items can cover a part of user demands. The hot events are used as one path of influence factors of the recall layer, so that the overall recommendation effect can be improved.

Recall for user search terms. Since the user may not know what the transaction name is before transacting business transaction, the transaction key is typically searched for priority to find the transaction entry. The keywords are matched to the office according to word segmentation techniques that utilize natural language processing of the user search words. The search items are directly related to the items possibly needing to be handled by the user, and the search items are used as one path of influence factors of the recall layer, so that the overall recommendation effect can be improved.

Recall of the user consultation event. The user consulting matters need to submit the consulting title and the consulting contents, and the user consulting matters indicate that the matters are more likely to need to be handled. The keywords are matched to the office according to word segmentation techniques of the user consultation words using natural language processing. Because the consultation items are directly related to the items possibly needing to be handled by the user, the overall recommendation effect can be improved by taking the consultation items as one influence factor of the recall layer.

Recall of user-based question-answering system events. Since the government system APP may be configured with an intelligent customer service (such as a conversation robot), instead of a manual customer service, keywords are matched to office matters by using a word segmentation technique of natural language processing according to the problem input by the user in the dialogue of the question-answering system. Because the matters mentioned by the user problems in the dialogue system are directly related to matters possibly needing to be handled by the user, the matters mentioned by the question-answering system are used as one influence factor of the recall layer, and the overall recommendation effect can be improved.

In operation S303, determining a recommendation score for each first recommendation object information is repeated: for each piece of first recommended object information, determining a recommended object association feature, and processing the recommended object association feature by using the first object recommendation model to determine a recommendation score of the first recommended object information.

In this embodiment, the first recommendation object information set with relatively comprehensive coverage is obtained after multi-channel recall. At this time, for each first recommended object information in the first recommended object information set, a score may be respectively made for each first recommended object information based on the history data of the current user or the history data of all users of the application, so as to predict the possibility that each first recommended object information is operated by the current user, so that whether to make a recommendation is determined based on the size of the possibility (which may be characterized based on the level of the recommendation score).

The first object recommendation model includes, but is not limited to, at least one of: a model based on a Boosting method (Boosting) algorithm, a model based on a gradient Boosting iterative decision tree (Gradient Boosting Decision Tree, GBDT) algorithm, and the like.

In operation S305, a second recommended object information set is determined based on the recommendation score of each of the first recommended object information to output the second recommended object information set.

For example, items output by each recall channel are scored using an object recommendation model, and the top N (TopN) item is finally selected. Where N may be based on user requirements or application scenarios, etc. For example, in a government application, a display area for displaying 6 or 11 items may be provided, and if the number of the output second recommended object information is less than 6 or 11 items, etc., the display area may be wasted, and if the number of the output second recommended object information is more than 11 items, it is inconvenient to display excessive second recommended object information, and thus the number of the second recommended object information may be equal to or slightly more than the number of objects that can be displayed in the display area.

According to the object recommending method provided by the embodiment of the disclosure, the multi-way recall and ordering structure is used for recommending government matters. Besides the collaborative filtering algorithm, a rule statistical algorithm based on user behavior weight is used, and user search recommendation, user consultation recommendation, user question and answer recommendation and popular items are combined in different proportions to generate recall layer recommendation items, and an optimal model is trained by using a sequencing algorithm to perform sequencing recommendation, so that the object recommendation accuracy under cold start and data sparse scenes is effectively improved.

In one embodiment, the second recommended object information set includes N pieces of second recommended object information, N being a positive integer greater than or equal to 2.

Accordingly, the method may further include the following operations. After the first recommendation object information set is obtained, P pieces of first candidate recommendation object information are determined from the first recommendation object information set to determine recommendation scores for each piece of first candidate recommendation object information, wherein P is a positive integer greater than N. The calculation amount of the first object recommendation model can be effectively reduced through the rule, and the processing efficiency is improved.

For example, if M recall channels each output the most O items, the M recall channels output the most p=m×o items of the first recommended object information. In order to improve the processing efficiency and reduce the resource occupancy rate, part of the first recommendation object information can be screened out from the first recommendation object information of the m×o item through a preset rule and the like. For example, 11 items are finally recommended, the first recommended object information set includes 250 items, and the first candidate recommended object information includes 30 items.

In one embodiment, the first candidate recommended object information includes first recommended object information respectively determined by a search item recall channel, a consultation item recall channel, and a question and answer system item recall channel, a sum of a first number and a second number is less than P, a third number is a number of bits of P, the first number is a number of first object information determined by a collaborative filtering recall channel, the second number is a number of first object information determined by a user behavior weight statistics recall channel, and the third number is a number of first object information determined by a popular item recall channel.

Fig. 5 schematically illustrates a structural diagram of first candidate recommended object information according to an embodiment of the present disclosure. As shown in fig. 5, different proportions can be set for the service data and the effect of single-path recall, for example, N items are finally required to be output in a recommended scene, and as shown in fig. 4 and 5, the 1 st and 2 nd paths (collaborative filtering recall channel and user behavior weight statistics recall channel) respectively output first recommendation object information of a first designated number and a second designated number (for example, N items); for the matters of searching, consulting and asking and answering of the user every day in the 3 rd, 4 th and 5 th paths, the number of the matters corresponding to the searching, consulting and asking and answering actions of the actual user in the same day is not large, and the matters are more concerned by the user, so that the number of each path is set as all matters of searching, consulting and asking and answering in the same day. And taking the hot item of the 6 th path as a recall layer bit filling item, wherein the total output number is Q item, and taking the hot item as a bit filling if the total output result of other paths is less than Q. If the 1 st path and the 2 nd path output 12 items, the 3 rd path, the 4 th path and the 5 th path output 1, 3 rd path and 2 nd path respectively, and the P first candidate recommended object information is 30 items, the 6 th path needs to be complemented with 30-12-12-1-1-2 = 2 items. The values of N, O, P, Q are merely exemplary, and may be values set based on predetermined rules, experience, or the like.

Through the combined recommendation of the plurality of recall channels, the problems of cold start and data sparseness which cannot be solved by using collaborative filtering algorithm recommendation alone are avoided. For a cold start scenario, new user logins may be recommended based on hot recommendations. The multi-way recall design can be supplemented according to historical behavior weight statistical rules of users, user searching, consultation and trending matters besides collaborative filtering recommendation, and the accuracy of overall system recommendation is improved.

In one embodiment, the method may further comprise the following operations. And adjusting the second recommended object information set based on the business strategy to output the adjusted second recommended object information set. Traffic policy adjustments such as setting whitelists, blacklists, etc. may be made. If the business needs to adjust the white list items, promoting some items, and adjusting the output result of the ordering layer.

In one embodiment, the input of the first object recommendation model may include user characteristics, object characteristics, and user behavior characteristics. The output of the first object recommendation model includes a probability that the first recommendation object information is operated within a specified time period after a specified time. The first object recommendation model may be a model for making object recommendation in various related technologies, such as a promotion tree model, a decision tree model, a neural network, and the like.

The sorting algorithm of the first object recommendation model constructs corresponding recommendation object association features based on the recalled item range, for example, based on clicking behaviors of a user, constructs the item heat clicked by the user, the total number of times the item is clicked, the number of times the item is clicked in the current month, values of three time periods in the last and last ten days, the total number of times the item is clicked in the current day, the number of times counted in the three time periods and the like.

FIG. 6 schematically illustrates a flow chart of determining a second set of recommended object information according to an embodiment of the disclosure.

As shown in fig. 6, the object recommendation model needs to perform feature extraction on training data, then perform model training, and then verify model effect to form a model through multiple parameter adjustment training. And outputting probability values clicked by a user on each item after the items output by the recall layer are input into the model, and selecting TopN items as final output results after sequencing.

Taking an offline object recommendation model as an example for explanation, for each first recommendation object information of a login user at a certain historical moment, recommendation object association features corresponding to the first recommendation object information are extracted, and model parameters are adjusted based on a back propagation algorithm and the like so that the model outputs labeling information approaching to the first recommendation object information. For example, for a first recommended object information output by the collaborative filtering recall channel, the first recommended object information has labeling information (e.g., whether the user has operated the first recommended object within a predetermined time period after the login time (e.g., 1 hour, 6 hours, 1 day, 1 week, 1 month, etc.).

FIG. 7 schematically illustrates a schematic diagram of a recommended object association feature according to an embodiment of the disclosure.

As shown in fig. 7, for example, based on the clicking behavior of a user, the heat of the item clicked by the user, the total number of times the item was clicked, the number of times the item was clicked in the current month, the values of three time periods in the last, middle and last ten days, the total number of times the item was clicked in the day and the number of times counted in the three time periods are extracted, and the item parent item id; the age, sex, region, liveness and other characteristics of the user; item id of user search, consultation, question and answer, etc.

For example, 22 feature values may be constructed as input features of the object recommendation model. The object recommendation model can select a machine learning method based on integrated learning, and can predict the probability value of a first recommendation object clicked by a user for the next day, so that the first recommendation object information can be ordered according to the size of the probability value, topN second recommendation objects are output, and TopN second recommendation objects are recommended.

The object recommendation model may use an XGBoost model, which is a lifting tree model, belonging to one of the lifting method (Boosting) algorithms, the idea of which is to integrate many weak classifiers together to form one strong classifier. After tuning using the XGBoost model, the output recall was approximately 83.43%. The objective function is composed of two parts, the first part is the gradient lifting algorithm loss, and the second part is the regularization term. The loss function is shown in the following equation.

Where n is the number of training function samples, 1 is the loss of a single sample, assuming it is a convex function, y _i ' is the predicted value of the model on the training sample, y _i Is the true tag value of the training sample.

The regularization term defines the complexity of the model as shown in equation (1).

Wherein, gamma and lambda are manually set parameters, omega is a vector formed by all leaf node values of the decision tree, and T is the number of leaf nodes.

It should be noted that, each recall algorithm in the multiple recalls may be replaced by other machine learning or deep learning algorithms, and the sorting algorithm may be replaced by other sorting algorithms, which is not limited herein.

In addition, the object recommendation method can also realize real-time recommendation through off-line calculation and near real-time calculation. That is, the object recommendation method may be applicable to real-time recommendation on-line.

In one embodiment, the method may further comprise the following operations.

In response to user behavior for the object from the client, the server side updates the first set of recommended object information.

Then, the following operations are repeated to determine a recommendation score for each of the updated first recommendation object information in the first recommendation object information set: for each piece of first recommended object information, determining a recommended object association characteristic, processing the recommended object association characteristic by using a second object recommendation model, and determining a recommendation score of the first recommended object information. The second object recommendation model may be a model suitable for the link framework, or other models capable of meeting the online prediction requirements.

In order to improve the response speed, the use method of the collaborative filtering algorithm of the collaborative filtering recall channel can be changed. For example, determining at least one first recommended object information based on the collaborative filtering algorithm may include the following operations.

First, a matrix of collaborative filtering algorithms is updated based on user behavior from clients for objects. At least one first current recommendation object information is then determined based on the updated matrix.

Specifically, the method can be realized based on a link framework, and based on offline recommendation calculation, single click behaviors are overlapped, the output of each recall channel is updated, and a second object recommendation model (such as GBDT based on a ranking algorithm) is used for calculating a final recommendation result.

The collaborative filtering recall channel is exemplified. The matrix calculated by the collaborative filtering algorithm based on the historical data of the previous day is shown in table 1.

TABLE 1

	Public service	Accumulation of money	Epidemic situation prevention and control	……
					User 1	a1	a2	a3	……
User 2	b1	b2	b3	……
					……	……	……	……	……

The user 2 performs a click operation on the epidemic situation prevention and control related content on the next day, and then the matrix of the collaborative filtering algorithm is updated as shown in table 2.

TABLE 2

	Public service	Accumulation of money	Epidemic situation prevention and control	……
					User 1	a1	a2	a3	……
User 2	b1	b2	b3’	……
					……	……	……	……	……

Wherein, the contents except b3' in the table 2 do not need to be recalculated, so that the calculation time of the collaborative filtering algorithm is effectively reduced.

According to the object recommendation method provided by the embodiment of the disclosure, the range of the more comprehensive recommended object is determined through multi-way recall under a cold start or data sparse scene (such as government class), and the recommendation accuracy is improved by combining an object recommendation model. The diversity of the finally obtained recommended results can be improved through multi-way recall, and scenes such as cold start can be dealt with. Through verification by using the same data, in a cold start or data sparse scene, the final recommendation recall rate is recommended to be about 25% only by using a collaborative filtering algorithm, and the recall rate of a final recommendation result is improved to be about 73% by using the object recommendation method provided by the embodiment of the disclosure.

Fig. 8 schematically illustrates a technical framework of an object recommendation method according to an embodiment of the present disclosure.

As shown in FIG. 8, the technical framework may be adapted for offline recommendation and online recommendation. The whole technical framework may include three layers: a recommendation layer, a calculation layer and a data layer.

The data layer is used for processing user attributes and user behavior data such as clicking, submitting, searching, consulting, question and answer behaviors and event attributes.

The recommendation layer obtains recommendation results through three processes of multi-way recall, recommendation ordering and rearrangement. The recommendation algorithm used by the recommendation layer not only comprises a collaborative filtering algorithm, but also considers data of other behaviors of the user, and after the multi-path recall result is synthesized, the ranking algorithm is called based on the constructed multi-dimensional user, behavior and item characteristics, and the recommendation result is output. For example, each of the multiple recalls outputs at most N items, the multiple recalls integrally outputs at most M items to be recommended, M is a value far smaller than the total number P of items, if the final recommendation is required to output 11 items, the total of the multiple recalls is 250 items to be recommended, and the multiple recalls integrally outputs only 30 items. And scoring and sorting the M items output by the recall layer by using a first object recommendation model for recommendation sorting, and finally selecting the previous TopN items. The third layer is rearrangement, and the TopN items are adjusted based on white lists, black lists and the like.

Fig. 9 schematically illustrates a block diagram of an object recommendation apparatus according to an embodiment of the present disclosure.

As shown in fig. 9, an object recommendation apparatus 900 of an embodiment of the present disclosure. The object recommendation device 900 may include a recall module 910, a recommendation score determination module 920, and a recommendation object determination module 930.

The recall module 910 is configured to determine at least one first recommended object information through M recall channels, respectively, to obtain a first recommended object information set, where M is a positive integer greater than or equal to 2.

The recommendation score determining module 920 is configured to determine a recommendation score of each first recommendation object information by repeating the following operations: for each piece of first recommended object information, determining a recommended object association feature, and processing the recommended object association feature by using the first object recommendation model to determine a recommendation score of the first recommended object information.

The recommended object determining module 930 is configured to determine a second recommended object information set based on the recommendation score of each of the first recommended object information to output the second recommended object information set.

It should be noted that the operations that the recall module 910, the recommendation score determining module 920, and the recommendation object determining module 930 may be executed by the above method may be the same as the relevant part of the content of the above method, which is not described herein again.

Any number of modules, sub-modules, units, sub-units, or at least some of the functionality of any number of the sub-units according to embodiments of the present disclosure may be implemented in one module. Any one or more of the modules, sub-modules, units, sub-units according to embodiments of the present disclosure may be implemented as split into multiple modules. Any one or more of the modules, sub-modules, units, sub-units according to embodiments of the present disclosure may be implemented at least in part as a hardware circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system-on-chip, a system-on-substrate, a system-on-package, an Application Specific Integrated Circuit (ASIC), or in any other reasonable manner of hardware or firmware that integrates or encapsulates the circuit, or in any one of or a suitable combination of three of software, hardware, and firmware. Alternatively, one or more of the modules, sub-modules, units, sub-units according to embodiments of the present disclosure may be at least partially implemented as computer program modules, which when executed, may perform the corresponding functions.

For example, any number of recall module 910, recommendation score determination module 920, and recommendation object determination module 930 may be combined in one module/unit/sub-unit or any number of modules/units/sub-units may be split into multiple modules/units/sub-units. Alternatively, at least some of the functionality of one or more of these modules/units/sub-units may be combined with at least some of the functionality of other modules/units/sub-units and implemented in one module/unit/sub-unit. At least one of recall module 910, recommendation score determination module 920, recommendation object determination module 930 may be implemented, at least in part, as hardware circuitry, such as a Field Programmable Gate Array (FPGA), programmable Logic Array (PLA), system on a chip, system on a substrate, system on a package, application Specific Integrated Circuit (ASIC), or in hardware or firmware, in any other reasonable manner of integrating or packaging circuitry, or in any one of or a suitable combination of three of software, hardware, and firmware, in accordance with embodiments of the present disclosure. Alternatively, at least one of recall module 910, recommendation score determination module 920, and recommendation object determination module 930 may be implemented at least in part as a computer program module that, when executed, performs the corresponding functions.

Fig. 10 schematically illustrates a block diagram of an electronic device adapted to perform an object recommendation method according to an embodiment of the present disclosure. The electronic device shown in fig. 10 is merely an example and should not be construed to limit the functionality and scope of use of the disclosed embodiments.

As shown in fig. 10, an electronic device 1000 according to an embodiment of the present disclosure includes a processor 1001 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 1002 or a program loaded from a storage section 1008 into a Random Access Memory (RAM) 1003. The processor 1001 may include, for example, a general purpose microprocessor (e.g., a CPU), an instruction set processor and/or an associated chipset and/or a special purpose microprocessor (e.g., an Application Specific Integrated Circuit (ASIC)), or the like. The processor 1001 may also include on-board memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of the method flows according to embodiments of the present disclosure.

In the RAM 1003, various programs and data necessary for the operation of the electronic apparatus 1000 are stored. The processor 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. The processor 1001 performs various operations of the method flow according to the embodiment of the present disclosure by executing programs in the ROM 1002 and/or the RAM 1003. Note that the program may be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in one or more memories.

According to an embodiment of the disclosure, the electronic device 1000 may also include an input/output (I/O) interface 1005, the input/output (I/O) interface 1005 also being connected to the bus 1004. The electronic device 1000 may also include one or more of the following components connected to the I/O interface 1005: an input section 1006 including a keyboard, a mouse, and the like; an output portion 1007 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc., and a speaker, etc.; a storage portion 1008 including a hard disk or the like; and a communication section 1009 including a network interface card such as a LAN card, a modem, or the like. The communication section 1009 performs communication processing via a network such as the internet. The drive 1010 is also connected to the I/O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like, is installed as needed in the drive 1010, so that a computer program read out therefrom is installed as needed in the storage section 1008.

According to embodiments of the present disclosure, a method flow according to embodiments of the present disclosure may be implemented as a computer program. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable storage medium, the computer program comprising program code for performing the method shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via the communication portion 1009, and/or installed from the removable medium 1011. The above-described functions defined in the electronic device of the embodiments of the present disclosure are performed when the computer program is executed by the processor 1001. According to embodiments of the present disclosure, the above-described electronic devices, apparatuses, means, modules, units, etc. may be implemented by computer program modules.

The present disclosure also provides a computer-readable storage medium that may be embodied in the apparatus/device/system described in the above embodiments; or may exist alone without being assembled into the apparatus/device/system. The computer-readable storage medium carries one or more programs which, when executed, implement methods in accordance with embodiments of the present disclosure.

According to embodiments of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example, but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this disclosure, a computer-readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the present disclosure, the computer-readable storage medium may include ROM 1002 and/or RAM 1003 and/or one or more memories other than ROM 1002 and RAM 1003 described above.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Those skilled in the art will appreciate that the features recited in the various embodiments of the disclosure and/or in the claims may be provided in a variety of combinations and/or combinations, even if such combinations or combinations are not explicitly recited in the disclosure. In particular, the features recited in the various embodiments of the present disclosure and/or the claims may be variously combined and/or combined without departing from the spirit and teachings of the present disclosure. All such combinations and/or combinations fall within the scope of the present disclosure.

The embodiments of the present disclosure are described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the embodiments cannot be used advantageously in combination. The scope of the disclosure is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be made by those skilled in the art without departing from the scope of the disclosure, and such alternatives and modifications are intended to fall within the scope of the disclosure.

Claims

1. An object recommendation method, comprising:

determining at least one first recommended object information of each through M recall channels respectively to obtain a first recommended object information set, wherein M is a positive integer greater than or equal to 2, and the recall channels comprise collaborative filtering recall channels, user behavior weight statistics recall channels and at least one of the following: a search item recall channel, a consultation item recall channel, a question and answer system item recall channel, or a hot item recall channel;

determining a recommendation score for each first recommendation object information in the first recommendation object information set by repeating the operations of: determining a recommendation object associated feature for each piece of first recommendation object information, and processing the recommendation object associated feature with a first object recommendation model to determine a recommendation score for the first recommendation object information; and

A second recommended object information set is determined based on the recommendation score of each piece of the first recommended object information to output the second recommended object information set.

2. The method of claim 1, wherein the second set of recommended object information includes N pieces of second recommended object information, N being a positive integer greater than or equal to 2;

the method further comprises the steps of: after said obtaining the first set of recommended object information,

p first candidate recommended object information is determined from the first recommended object information set to determine a recommended score for each first candidate recommended object information, and P is a positive integer greater than N.

3. The method of claim 2, wherein the first candidate recommended object information includes first recommended object information determined by each of the search event recall channel, consultation event recall channel, and question and answer system event recall channel, a sum of a first number and a second number being less than P, a third number being a complement number of P, the first number being a number of first object information determined by the collaborative filtering recall channel, the second number being a number of first object information determined by the user behavior weight statistics recall channel, the third number being a number of first object information determined by the trending event recall channel.

4. The method of claim 1, wherein the determining, via the M recall channels, the respective at least one first recommended object information comprises:

determining at least one first recommended object information based on a collaborative filtering algorithm for the collaborative filtering recall channel;

and for the user behavior weight statistics recall channel, respectively executing the following operations for each user: counting the behaviors of the user aiming at each object information respectively based on the behavior weights to obtain the statistical scores of the user aiming at each object information respectively, and determining at least one first recommended object information based on the statistical scores;

for at least one channel of the consultation item recall channel, the question and answer system item recall channel or the popular item recall channel, determining at least one first recommended object information of each channel by carrying out keyword matching on word segmentation results of input information; and

and regarding the hot event recall channel, taking the first Q pieces of the hot events counted in a specified period as at least one piece of first recommended object information, wherein Q is a positive integer greater than or equal to zero.

5. The method of claim 4, further comprising:

In response to user behavior for an object from a client, the server updates the first recommended object information set; and

determining a recommendation score for each of the updated first recommendation object information in the set of first recommendation object information by repeating the operations of: and determining a recommendation object association characteristic for each piece of first recommendation object information, processing the recommendation object association characteristic by using a second object recommendation model, and determining a recommendation score of the first recommendation object information, wherein the second object recommendation model is applicable to a furin frame.

6. The method of claim 5, wherein the determining at least one first recommended object information based on a collaborative filtering algorithm comprises:

updating a matrix of a collaborative filtering algorithm based on the user behavior for the object from the client; and

at least one first current recommended object information is determined based on the updated matrix.

7. The method of any one of claims 1 to 6, wherein:

the input of the first object recommendation model comprises user characteristics, object characteristics and user behavior characteristics; and

the output of the first object recommendation model includes a probability that the first recommendation object information is operated within a specified time period after a specified time.

8. The method of any one of claims 1 to 6, further comprising:

and adjusting the second recommended object information set based on the business strategy to output an adjusted second recommended object information set.

9. An object recommendation device, comprising:

the recall module is used for respectively determining at least one first recommended object information through M recall channels to obtain a first recommended object information set, M is a positive integer greater than or equal to 2, and the recall channels comprise collaborative filtering recall channels, user behavior weight statistics recall channels and at least one of the following: a search item recall channel, a consultation item recall channel, a question and answer system item recall channel, or a hot item recall channel;

a recommendation score determining module for determining a recommendation score of each first recommendation object information by repeating the following operations: for each piece of first recommended object information, determining a recommended object association feature, and processing the recommended object association feature by using a first object recommendation model to determine a recommendation score of the first recommended object information; and

and the recommendation object determining module is used for determining a second recommendation object information set based on the recommendation score of each piece of first recommendation object information so as to output the second recommendation object information set.

10. An electronic device, comprising:

one or more processors; and

storage means for storing one or more programs,

wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-8.

11. A computer readable storage medium having stored thereon executable instructions which when executed by a processor cause the processor to implement the method of any of claims 1 to 8.