WO2022218181A1

WO2022218181A1 - Information recommendation method and device and computer storable medium

Info

Publication number: WO2022218181A1
Application number: PCT/CN2022/085053
Authority: WO
Inventors: 张青青; 李山林; 余聪
Original assignee: 北京沃东天骏信息技术有限公司; 北京京东世纪贸易有限公司
Priority date: 2021-04-15
Filing date: 2022-04-02
Publication date: 2022-10-20
Also published as: CN113111182A

Abstract

Provided are an information recommendation method and device and a computer storable medium, relating to the technical field of computers. The information recommendation method comprises: determining at least one target keyword and a target vector thereof according to a keyword to be processed (S10); for each target keyword, obtaining, from historical data corresponding to an object related to said keyword, multiple reference keywords that belong to the same keyword category as each target keyword (S20); determining, according to multiple reference vectors corresponding to the multiple reference keywords, a comprehensive vector corresponding to the object (S30); and recommending, according to the similarity between the comprehensive vector and each target vector, information related to said keyword (S40). According to the present invention, the accuracy of recommendation can be improved.

Description

Information recommendation method and device, and computer storable medium

CROSS-REFERENCE TO RELATED APPLICATIONS

The present application is based on the CN application number 202110405818.1 and the filing date is April 15, 2021, and claims its priority. The disclosure of the CN application is hereby incorporated into the present application as a whole.

technical field

The present disclosure relates to the field of computer technology, and in particular, to a method and apparatus for recommending information, and a computer-storable medium.

Background technique

With the rapid development of the live broadcast industry, how to accurately screen users with goods for live broadcast has become the key to promoting the development of the live broadcast industry.

In the related art, for an item to be delivered, by counting the frequency of delivery and the sales volume of different delivery users, a seller with a high frequency of delivery and a large volume of delivery is recommended to the merchant.

SUMMARY OF THE INVENTION

According to a first aspect of the present disclosure, an information recommendation method is provided, comprising: determining at least one target keyword and its target vector according to the keywords to be processed, and the target keyword and the target vector are in one-to-one correspondence; for each target key word, obtain a plurality of reference keywords belonging to the same keyword category as each target keyword from the historical data corresponding to the object related to the keyword to be processed; multiple reference vectors of Keyword related information.

In some embodiments, determining the comprehensive vector corresponding to the object includes: for the object, determining conversion data corresponding to each reference keyword according to corresponding historical data; according to the plurality of reference vectors and Its corresponding transformation data, the integrated vector is determined.

In some embodiments, determining the comprehensive vector according to the multiple reference vectors and their corresponding conversion data includes: determining a corresponding conversion weight according to the conversion data corresponding to each reference keyword; using multiple conversion weights , performing a weighting operation on the multiple reference vectors to obtain the integrated vector.

In some embodiments, the object is a user with goods, the historical data is historical data with goods, and the historical data with goods of the user with goods includes daily data in a historical time period, and the daily data is The product data includes at least one of the click volume, order volume, click follower volume and online follower volume of the carrying product corresponding to each reference keyword in one day, and the click follower volume is the user carrying the product within one day. The number of fans who clicked on the goods with the goods among the online fans of the The click volume is positively correlated and negatively correlated with the online fan volume, the conversion rate is positively correlated with the order volume and negatively correlated with the click volume, and the fan activity rate is the click volume and the online fan volume ratio between.

In some embodiments, there are multiple users with goods, the click-through rate is positively correlated with the average click volume and negatively correlated with the average online fan volume, and the average click volume is the average click volume of the multiple users with goods. The average of clicks, the average online fans is the average of online fans of the multiple users with goods; the conversion rate is positively correlated with the average order volume and negatively correlated with the average clicks, so The average order quantity is the average of the order quantities of the multiple users with goods.

In some embodiments, determining the corresponding conversion weight includes: determining the average of the click-through rate, the conversion rate, and the fan activity rate as the conversion weight.

In some embodiments, the historical data corresponding to the object includes at least one group of order data derived from the object, and each group of order data includes item identifiers of a plurality of items, from the data corresponding to the object related to the keyword to be processed. In the historical data, acquiring a plurality of reference keywords belonging to the same keyword category as each target keyword includes: determining a plurality of item groups according to each group of order data derived from the object, and each item group includes two Item identifiers of different items; according to each group of order data derived from the object, determine the association weight value between the two items in each item group corresponding to the item identifier. The stronger the correlation is; the item identifiers in the multiple item groups are used as vertices, and the associated weights are used as edges to construct a graph model; using the graph model, from the historical data corresponding to the object, obtain the Describe multiple reference keywords that each target keyword belongs to the same keyword category.

In some embodiments, the order data further includes other order parameters corresponding to the item identifier, and the graph model is used to obtain the same key as each target keyword from historical data corresponding to the object. The multiple reference keywords of the word category include: using a random walk algorithm to randomly walk the graph model until the walking stop condition is met, and at least one item identifier is obtained as a reference item identifier; for each target Keywords, from each group of order data derived from the object, select the item ID or other order parameters that belong to the same keyword category as each target keyword and correspond to the reference item ID, as the reference keyword .

In some embodiments, the order data further includes an order date, and determining an association weight value between two items in each item group corresponding to the item identifiers includes: The difference between the order dates determines the association weight value between the two items in each item group corresponding to the item identifier, and the association weight value is negatively correlated with the difference between the order dates.

In some embodiments, the order data further includes an order date, and according to the order data corresponding to each user with goods, determining a plurality of groups of goods with goods includes: according to each group of order data, determining at least one set of items, each set of goods Including multiple item IDs arranged in order of order, the difference between the latest order date and the earliest order date corresponding to multiple item IDs in the same item set is less than the date threshold; for each item set, the Two adjacent item identifiers are divided into one item group to obtain the plurality of item groups.

In some embodiments, recommending information related to the to-be-processed keyword includes: when the object is a plurality of objects, selecting an object to be recommended from the plurality of objects, and recommending the to-be-recommended object information about the object.

In some embodiments, when the object is one object, the at least one target keyword includes multiple target keywords, and recommending information related to the keyword to be processed includes: when the object is one In the case of the target keyword, the keyword to be recommended is selected from the multiple target keywords, and the related information of the keyword to be recommended is recommended.

In some embodiments, determining at least one target keyword according to the keyword to be processed includes: converting the keyword to be processed into at least one target keyword according to a keyword category corresponding to the keyword to be processed.

In some embodiments, when the object is a user with goods, the keyword categories include user identification categories, item identification categories, brand identification categories, store identification categories, and categories.

According to a second aspect of the present disclosure, there is provided an information recommendation device, comprising: a first determination module configured to determine at least one target keyword and its target vector according to the keywords to be processed, the target keyword and the target vector one by one Corresponding; the acquisition module is configured to, for each target keyword, acquire a plurality of references belonging to the same keyword category with each target keyword from historical data corresponding to the object related to the keyword to be processed keywords; a second determining module, configured to determine a comprehensive vector corresponding to the object according to a plurality of reference vectors corresponding to the plurality of reference keywords, and the reference keywords and the reference vectors are in one-to-one correspondence; a recommendation module, is configured to recommend information related to the keyword to be processed according to the similarity between the integrated vector and each of the target vectors.

According to a third aspect of the present disclosure, there is provided an information recommendation apparatus, comprising: a memory; and a processor coupled to the memory, the processor configured to execute any one of the above based on instructions stored in the memory The information recommendation method described in the embodiment.

According to a fourth aspect of the present disclosure, there is provided a computer-storable medium on which computer program instructions are stored, and when the instructions are executed by a processor, implement the information recommendation method described in any of the foregoing embodiments.

Description of drawings

The accompanying drawings, which form a part of the specification, illustrate embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure.

The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, wherein:

FIG. 1 is a flowchart illustrating an information recommendation method according to some embodiments of the present disclosure;

FIG. 2 is a flowchart illustrating obtaining a plurality of reference keywords belonging to the same keyword category as each target keyword according to some embodiments of the present disclosure;

3 is a flow diagram illustrating determining a synthesis vector corresponding to an object according to some embodiments of the present disclosure;

4 is a block diagram illustrating an information recommendation apparatus according to some embodiments of the present disclosure;

FIG. 5 is a block diagram illustrating an information recommendation apparatus according to other embodiments of the present disclosure;

6 is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure.

Detailed ways

Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

Meanwhile, it should be understood that, for the convenience of description, the dimensions of various parts shown in the accompanying drawings are not drawn in an actual proportional relationship.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application or uses in any way.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification.

In all examples shown and discussed herein, any specific value should be construed as illustrative only and not as limiting. Accordingly, other examples of exemplary embodiments may have different values.

It should be noted that like numerals and letters refer to like items in the following figures, so once an item is defined in one figure, it does not require further discussion in subsequent figures.

In the related art, the matching degree between the user carrying the goods and the goods to be carried is not considered, so the accuracy of the recommendation is poor.

In view of the above technical problems, the present disclosure proposes an information recommendation method, which can improve the accuracy of recommendation.

FIG. 1 is a flowchart illustrating an information recommendation method according to some embodiments of the present disclosure.

As shown in Figure 1, the information recommendation method includes: step S10, determining at least one target keyword and its target vector according to the keywords to be processed; step S20, for each target keyword, from the object related to the keyword to be processed In the corresponding historical data, obtain a plurality of reference keywords belonging to the same keyword category as each target keyword; Step S30, according to a plurality of reference vectors corresponding to the plurality of reference keywords, determine a comprehensive vector corresponding to the object; Step S40, recommend information related to the keyword to be processed according to the similarity between the comprehensive vector and each target vector. In some embodiments, the information recommendation method is performed by an information recommendation apparatus. The information recommendation method can be applied to the scene of live broadcast with goods, and can also be applied to the scene of advertising media selection.

In step S10, at least one target keyword and its target vector are determined according to the keywords to be processed. Target keywords and target vectors correspond one-to-one. In some embodiments, the keyword to be processed may be a search term input by a search user.

In some embodiments, for example, the keywords to be processed include, but are not limited to, the item identification of the carried item, the user identification of the carried user, the brand identification of the carried item, and the category identification of the carried item. For example, the item identifier can be an item ID (Identity Document, identification number) or an item name, the user identifier can be a user ID or a user name, the brand identifier can be a brand ID or a brand name, and the category identifier can be a category ID or category name.

In some embodiments, the keyword to be processed is converted into at least one target keyword according to the keyword category corresponding to the keyword to be processed. In some embodiments, when the object related to the keyword to be processed is a user with goods, the keyword category includes but is not limited to user identification category, item identification category, brand identification category, store identification category, and category category.

For example, when the keyword category corresponding to the keyword to be processed is the item identification category, the keyword to be processed itself is used as the target keyword.

For another example, in the case where the keyword category corresponding to the keyword to be processed is the user identification category, the keyword to be processed is converted into the item identification, category identification and brand identification corresponding to the user identification corresponding to the keyword to be processed. at least one of. In some embodiments, one user ID corresponds to multiple item IDs, multiple category IDs, or multiple brand IDs, then the target vector corresponding to the target keyword formed by multiple item IDs may be multiple IDs of multiple item IDs Average or weighted average of vectors. Similarly, target vectors corresponding to target keywords composed of multiple category identifiers or multiple brand identifiers can be determined.

For another example, in the case where the keyword category corresponding to the keyword to be processed is a user identification category, the keyword to be processed may also be converted into item IDs of multiple candidate items.

In step S20, for each target keyword, multiple reference keywords belonging to the same keyword category as each target keyword are acquired from historical data corresponding to objects related to the keyword to be processed.

For example, the above-mentioned step S20 is implemented in the manner shown in FIG. 2 .

FIG. 2 is a flowchart illustrating obtaining a plurality of reference keywords belonging to the same keyword category as each target keyword according to some embodiments of the present disclosure.

As shown in FIG. 2 , acquiring a plurality of reference keywords belonging to the same keyword category as each target keyword includes steps S21 to S24. For example, the historical data corresponding to the above-mentioned object includes at least one group of order data derived from the object. Each set of order data includes item identifiers for multiple items. For example, when the object is a user with goods, each group of order data derived from the object is the order data of each fan of the user with goods. The multiple items included in each set of order data are multiple carry items.

For example, the order data of a fan A can be expressed as {[item ID=3645, date=20200101, brand ID=Jordan, shop=Jordan store, category=sneakers], [item ID=4543, date=20200101, brand ID=Nike, Store=Nike Store, Category=Sneakers], [Item ID=712345, Date=20200102, Brand ID=Huawei, Store=Huawei Store, Category=Mobile], [Item ID=546574, Date= 20200108, Brand ID = Picardine, Store = Picardine, Category = Jeans]}.

In step S21, a plurality of item groups are determined according to each group of order data derived from the object. Each item group includes item IDs for two different items. Taking the order data including the order date as an example, according to each group of order data, at least one item set is determined, and each item set includes multiple item identifiers arranged in the order of the order, and the multiple item identifiers in the same item set correspond to The difference between the latest order date and the earliest order date is less than a date threshold (eg, 3 days). For each item set, two adjacent item identifiers are divided into an item group to obtain multiple item groups.

For example, the order data of fan A can be divided into order set B and order set C. Order set B is {[Item ID=3645, Date=20200101, Brand ID=Jordan, Store=Jordan Store, Category=Sneakers], [Item ID=4543, Date=20200101, Brand ID=Nike, Store=Nike Store, Category=Sneakers], [Item ID=712345, Date=20200102, Brand ID=Huawei, Store=Huawei Store, Category=Mobile]}. Order set C is {[ItemID=546574, Date=20200108, BrandID=Picardin, Shop=Picardin, Category=Jeans], [ItemID=546578, Date=20200109, BrandID=Picard Dan, shop = Picardine, category = denim jacket]}. The difference between the latest order date and the earliest order date in a single order collection is less than 3 days. According to the two order sets, two item sets can be determined, namely {commodity ID=3645, commodity ID=4543, commodity ID=712345}, {commodity ID=546574, commodity ID=546578}.

In step S22, according to each group of order data derived from the object, determine the association weight value between the two items corresponding to the item identifiers in each item group. The larger the association weight value, the stronger the correlation between the two items.

In some embodiments, the order data also includes the order date. According to the difference between the order dates corresponding to the two item identifiers in each item group, the association weight value between the two items corresponding to the item identifiers in each item group is determined. The association weight value is negatively correlated with the difference between the order dates.

For example, the associated weight value is

D is the difference between order dates (later order dates minus earlier order dates), and M is a preset positive integer. For example, M is set to 4. This is because the difference between order dates will not exceed 4.

In step S23, a graph model is constructed using the item identifiers in the multiple item groups as vertices and the associated weights as edges. For example, after obtaining the associated weight value, the order set B is processed, and the processed set {[item ID=3645, date=20200101, brand ID=Jordan, shop=Jordan store, category=sports shoes], [ItemID=4543, Date=20200101, BrandID=Nike, Store=Nike Store, Category=Sneakers], Weight=1}, and {[ItemID=4543, Date=20200101, BrandID=Nike, Store = Nike store, category = sports shoes], [item ID = 712345, date = 20200102, brand ID = Huawei, store = Huawei store, category = mobile], weight = 0.75}. From the processed collection, a graph model can be built.

In step S24, a graph model is used to acquire a plurality of reference keywords belonging to the same keyword category as each target keyword from the historical data corresponding to the object.

In some embodiments, the order data also includes other order parameters corresponding to the item identification. The above step S24 can be implemented in the following manner.

First, use the random walk algorithm to randomly walk the graph model until the stop condition of the walk is satisfied, and at least one item identifier is obtained as a reference item identifier. For example, the parameters of the random walk include the maximum step size of the walk (eg, 13), and the number of traversals per vertex (eg, 4). The stop condition of the walk is the total length of the walk. In some embodiments, the random walk algorithm is an Alias sampling based random walk algorithm.

Then, for each target keyword, from each group of order data derived from the object, select the item ID or other order parameters that belong to the same keyword category as each target keyword and correspond to the reference item ID as the reference keyword . Taking the order data including item ID, brand ID, and category ID as an example, in the case that the target keyword belongs to the item ID category, the reference keyword is also the item ID (ie, the reference item ID). In the case where the target keyword belongs to the brand identification category, the reference keyword is the brand identification corresponding to the reference item identification. In the case that the target keyword belongs to the category identification category, the reference keyword is the category identification corresponding to the reference item identification.

In the above embodiment, by calculating the correlation weight value and constructing a graph model, the correlation between items can be constructed, so that the reference keyword most related to the target keyword can be determined by using the correlation between the items, and the recommendation can be further improved. accuracy.

Returning to FIG. 1, in step S30, a comprehensive vector corresponding to an object is determined according to a plurality of reference vectors corresponding to a plurality of reference keywords. Taking the object as a user with goods as an example, each user with goods corresponds to a comprehensive vector.

For example, the above-mentioned step S30 is implemented in the manner shown in FIG. 3 .

Figure 3 is a flow diagram illustrating determining a synthesis vector corresponding to an object in accordance with some embodiments of the present disclosure.

As shown in FIG. 3 , determining the integrated vector corresponding to the object includes steps S31 to S32.

In step S31, the conversion data corresponding to each reference keyword is determined according to the historical data corresponding to the object. Conversion data is data that measures the ability to generate specific contributions (eg, traffic contributions such as click-through rate, conversion rate, fan activity rate, etc.) based on each reference keyword. In some embodiments, when the object is a user with goods, the historical data is historical delivery data, and the transformation data is delivery conversion data. In other embodiments, when the object is an advertising medium, the historical data is historical advertising data, and the conversion data is advertising conversion data.

In step S32, a comprehensive vector is determined according to a plurality of reference vectors and their corresponding transformation data. By combining the transformation data to determine the comprehensive vector, not only the matching degree between the keyword to be processed and the object, but also the transformation ability of the object is considered, which can further improve the accuracy of the recommendation.

In some embodiments, the corresponding conversion weight is determined according to the conversion data corresponding to each reference keyword. Using multiple transformation weights, weighting operations are performed on multiple reference vectors to obtain a comprehensive vector. By converting the conversion data into conversion weights and performing weighting operations on multiple reference vectors, the conversion ability of the object can be comprehensively evaluated, and the accuracy of the recommendation can be further improved. The transformation weight is the weight value of each reference vector.

Taking the object as the user with goods and the historical data as the historical delivery data as an example, the historical delivery data of the user with goods includes the daily delivery data in the historical time period. The daily delivery data includes at least one of the click volume, order volume, click followers volume and online followers volume of the delivered product corresponding to each reference keyword in one day. The number of clicked followers refers to the number of followers who clicked on the goods with goods among the online followers of the users who brought goods in one day.

Taking the object as the user with the goods and the historical data as the historical data of the goods as an example, the conversion data includes the daily click through rate (Click Through Rate, CTR), conversion rate (Conversion Rate, CVR) and fan activity rate in the historical time period. at least one of. Click-through rate is positively correlated with clicks and negatively correlated with online followers. Conversion rate is positively correlated with order volume and negatively correlated with click volume. The fan activity rate is the ratio between the number of clicked fans and the number of online fans.

Taking the conversion data including the daily click-through rate, conversion rate, and fan activity rate in a historical time period as an example, the average value of the click-through rate, conversion rate, and fan activity rate can be determined as the conversion weight corresponding to the reference keyword.

In some embodiments, when there are multiple users with goods, the click-through rate is positively correlated with the average click volume and negatively correlated with the average online fan count. The average click volume is the average of the click volume of multiple users with goods. The average number of online fans is the average of the online fans of multiple users with goods. Conversion rate is positively correlated with average order volume and negatively correlated with average click volume. The average order size is the average of the order sizes of multiple users with goods.

In the above embodiment, the conversion data is modified by using the average online followers, average clicks and average orders corresponding to multiple users with goods, that is, using the Bayesian smoothing idea to modify the conversion data, which can make the conversion data closer to the real value , so as to further improve the accuracy of the recommendation.

Returning to FIG. 1, in step S40, information related to the keywords to be processed is recommended according to the similarity between the comprehensive vector and each target vector. For example, the similarity is the vector inner product or cosine similarity between vectors.

In the case that the objects are multiple objects, the objects to be recommended are selected from the multiple objects, and the relevant information of the objects to be recommended is recommended. For example, a user to be recommended is selected from multiple users with goods, and user information of the user to be recommended is recommended. In this case, a user with goods matching the keyword to be processed is selected from multiple users with goods, so that the selected user with goods can carry goods for the item corresponding to the keyword to be processed. In some embodiments, the user information of the user with goods corresponding to the comprehensive vector with the largest sum of similarity of each target vector may be recommended.

When the object is one object, there are plural target keywords. From a plurality of target keywords, a keyword to be recommended is selected, and relevant information of the keyword to be recommended is recommended. For example, from a plurality of candidate items, an item to be recommended is selected, and item information of the item to be recommended is recommended. In this case, a matching item to be recommended is selected for a user with goods, so that the user with goods can carry goods for the item to be recommended. In some embodiments, the item information of the candidate item corresponding to the target vector with the greatest similarity with the comprehensive vector may be recommended.

In the above embodiment, the comprehensive vector of the object is determined by converting the keywords to be processed into the target keywords, and using the reference vectors of multiple reference keywords belonging to the same keyword category as the target keyword obtained based on historical data. , the recommendation is based on the similarity between the comprehensive vector and the target vector, and the matching degree between the object and the keyword to be processed is comprehensively considered, which can improve the accuracy of the recommendation.

FIG. 4 is a block diagram illustrating an information recommendation apparatus according to some embodiments of the present disclosure.

As shown in FIG. 4 , the information recommendation apparatus 4 includes a first determination module 41 , an acquisition module 42 , a second determination module 43 and a recommendation module 44 .

The first determination module 41 is configured to determine at least one target keyword and its target vector according to the keywords to be processed, for example, to perform step S10 as shown in FIG. 1 . Target keywords and target vectors correspond one-to-one.

The acquisition module 42 is configured to, for each target keyword, acquire a plurality of reference keywords belonging to the same keyword category with each target keyword from the historical data corresponding to the object related to the keyword to be processed, for example, perform the following steps: Step S20 shown in FIG. 1 .

The second determining module 43 is configured to determine a comprehensive vector corresponding to the object according to the plurality of reference vectors corresponding to the plurality of reference keywords, for example, performing step S30 shown in FIG. 1 for one-to-one correspondence between the reference keywords and the reference vectors.

The recommendation module 44 is configured to recommend information related to the keyword to be processed according to the similarity between the integrated vector and each target vector, for example, to perform step S40 as shown in FIG. 1 .

FIG. 5 is a block diagram illustrating an information recommendation apparatus according to other embodiments of the present disclosure.

As shown in FIG. 5 , the information recommendation apparatus 5 includes a memory 51 ; and a processor 52 coupled to the memory 51 . The memory 51 is used for storing instructions for executing the corresponding embodiments of the information recommendation method. The processor 52 is configured to execute the information recommendation method in any of some embodiments of the present disclosure based on the instructions stored in the memory 51 .

As shown in FIG. 6, computer system 60 may take the form of a general-purpose computing device. Computer system 60 includes memory 610, processor 620, and bus 600 that connects various system components.

Memory 610 may include, for example, system memory, non-volatile storage media, and the like. The system memory stores, for example, an operating system, an application program, a boot loader (Boot Loader), and other programs. System memory may include volatile storage media such as random access memory (RAM) and/or cache memory. The non-volatile storage medium stores, for example, instructions for performing the corresponding embodiments of at least one of the information recommendation methods. Non-volatile storage media include, but are not limited to, magnetic disk memory, optical memory, flash memory, and the like.

Processor 620 may be implemented as a general purpose processor, digital signal processor (DSP), application specific integrated circuit (ASIC), field programmable gate array (FPGA) or other programmable logic device, discrete hardware components such as discrete gates or transistors. accomplish. Correspondingly, each module such as the judging module and the determining module can be implemented by a central processing unit (CPU) running the instructions in the memory for executing the corresponding steps, or can be implemented by a dedicated circuit for executing the corresponding steps.

The bus 600 may use any of a variety of bus structures. For example, bus structures include, but are not limited to, Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus.

The computer system 60 may also include an input-output interface 630, a network interface 640, a storage interface 650, and the like. The

interfaces

630 , 640 , 650 and the memory 610 and the processor 620 can be connected through the bus 600 . The input and output interface 630 may provide a connection interface for input and output devices such as a monitor, a mouse, and a keyboard. Network interface 640 provides a connection interface for various networked devices. The storage interface 650 provides a connection interface for external storage devices such as a floppy disk, a U disk, and an SD card.

Aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks, can be implemented by computer readable program instructions.

These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer or other programmable device to produce a machine such that execution of the instructions by the processor produces one or more blocks in the flowchart and/or block diagrams device with the specified function.

Also stored in computer readable memory are these computer readable program instructions, which cause the computer to operate in a particular manner resulting in an article of manufacture including implementing the functions specified in one or more blocks of the flowchart and/or block diagrams instruction.

The present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.

Through the information recommendation method and device, and the computer-storable medium in the above-mentioned embodiments, the accuracy of recommendation can be improved.

So far, the information recommendation method and apparatus, and the computer-storable medium according to the present disclosure have been described in detail. Some details that are well known in the art are not described in order to avoid obscuring the concept of the present disclosure. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

Claims

An information recommendation method, including:

According to the keywords to be processed, at least one target keyword and its target vector are determined, and the target keyword and the target vector are in one-to-one correspondence;

For each target keyword, obtain a plurality of reference keywords belonging to the same keyword category as each target keyword from the historical data corresponding to the object related to the keyword to be processed;

According to a plurality of reference vectors corresponding to the plurality of reference keywords, a comprehensive vector corresponding to the object is determined, and the reference keywords and the reference vectors are in one-to-one correspondence;

According to the similarity between the comprehensive vector and each target vector, information related to the keyword to be processed is recommended.
The information recommendation method according to claim 1, wherein determining the comprehensive vector corresponding to the object comprises:

For the object, according to the corresponding historical data, determine the conversion data corresponding to each of the reference keywords;

The integrated vector is determined according to the plurality of reference vectors and their corresponding transformation data.
The information recommendation method according to claim 2, wherein determining the comprehensive vector according to the plurality of reference vectors and their corresponding transformation data comprises:

According to the conversion data corresponding to each reference keyword, determine the corresponding conversion weight;

Using a plurality of transformation weights, a weighting operation is performed on the plurality of reference vectors to obtain the comprehensive vector.
The information recommendation method according to claim 3, wherein the object is a user with goods, the historical data is historical data with goods, and the historical data with goods of the user with goods includes daily goods in a historical time period. Data, the daily delivery data includes at least one of the click volume, order volume, click fan volume and online fan volume of the delivered product corresponding to each reference keyword in one day, and the click volume is: The number of fans who click on the item with the item among the online fans of the user with the item in one day,

The conversion data includes at least one of click-through rate, conversion rate, and fan activity rate for each day in the historical time period, and the click-through rate is positively correlated with the click volume and negatively correlated with the online fan volume Correlation, the conversion rate is positively correlated with the order volume and negatively correlated with the click volume, and the fan activity rate is a ratio between the click volume and the online number of followers.
The information recommendation method according to claim 4, wherein there are multiple users with goods, the click-through rate is positively correlated with the average click volume and negatively correlated with the average online fan volume, and the average click volume is the The average value of the clicks of the multiple users with goods, and the average online fans is the average of the online fans of the multiple users with goods;

The conversion rate is positively correlated with the average order volume and negatively correlated with the average click volume, and the average order volume is an average value of the order volumes of the plurality of users with goods.
The information recommendation method according to claim 4, wherein determining the corresponding conversion weight comprises:

The average value of the click-through rate, the conversion rate, and the fan activity rate is determined as the conversion weight.
The information recommendation method according to claim 1, wherein the historical data corresponding to the object includes at least one group of order data derived from the object, and each group of order data includes item identifiers of a plurality of items, from which data are to be processed. In the historical data corresponding to the keyword-related objects, acquiring a plurality of reference keywords belonging to the same keyword category as each target keyword includes:

Determine a plurality of item groups according to each group of order data derived from the object, and each item group includes item identifiers of two different items;

According to each group of order data derived from the object, determine the association weight value between the two items corresponding to the item identifier in each item group, the larger the association weight value, the stronger the correlation between the two items;

Constructing a graph model with item identifiers in the plurality of item groups as vertices and associated weights as edges;

Using the graph model, a plurality of reference keywords belonging to the same keyword category as each target keyword are acquired from historical data corresponding to the object.
The information recommendation method according to claim 7, wherein the order data further includes other order parameters corresponding to the item identifiers, and the graph model is used to obtain information related to each item from historical data corresponding to the object by using the graph model. Multiple reference keywords belonging to the same keyword category for each target keyword include:

A random walk algorithm is used to randomly walk the graph model until the walking stop condition is met, and at least one item identifier is obtained as a reference item identifier;

For each target keyword, from each group of order data derived from the object, select an item ID or other order that belongs to the same keyword category as each target keyword and corresponds to the reference item ID parameters, as reference keywords.
The information recommendation method according to claim 7, wherein the order data further includes an order date, and determining the association weight value between the two items corresponding to the item identifiers in each item group comprises:

According to the difference between the order dates corresponding to the two item IDs in each item group, determine the association weight value between the two items corresponding to the item ID in each item group, the association weight value and the order date The difference between them is negatively correlated.
The information recommendation method according to claim 7, wherein the order data further includes an order date, and according to the order data corresponding to each user with goods, determining the plurality of groups of goods with goods includes:

According to each group of order data, at least one item set is determined, each item set includes multiple item identifiers arranged in order of order, and the latest order date and earliest order date corresponding to multiple item identifiers in the same item set The difference between is less than the date threshold;

For each item set, two adjacent item identifiers are divided into one item group to obtain the plurality of item groups.
The information recommendation method according to claim 1, wherein recommending the information related to the keyword to be processed comprises:

When the objects are multiple objects, an object to be recommended is selected from the multiple objects, and relevant information of the object to be recommended is recommended.
The information recommendation method according to claim 1, wherein, when the object is one object, the at least one target keyword includes a plurality of target keywords, and recommending information related to the keyword to be processed includes :

In the case that the object is one object, a keyword to be recommended is selected from the plurality of target keywords, and related information of the keyword to be recommended is recommended.
The information recommendation method according to claim 1, wherein, according to the keywords to be processed, determining at least one target keyword comprises:

Convert the keyword to be processed into at least one target keyword according to the keyword category corresponding to the keyword to be processed.
The information recommendation method according to claim 1, wherein, when the object is a user with goods, the keyword categories include user identification category, item identification category, brand identification category, store identification category and category category.
An information recommendation device, comprising:

The first determining module is configured to determine at least one target keyword and its target vector according to the keywords to be processed, and the target keywords and target vectors correspond one-to-one;

The acquisition module is configured to, for each target keyword, acquire a plurality of reference keywords belonging to the same keyword category as each target keyword from historical data corresponding to objects related to the keyword to be processed ;

The second determining module is configured to determine a comprehensive vector corresponding to the object according to a plurality of reference vectors corresponding to the plurality of reference keywords, and the reference keywords and the reference vectors are in one-to-one correspondence;

The recommendation module is configured to recommend information related to the keyword to be processed according to the similarity between the comprehensive vector and each target vector.
An information recommendation device, comprising:

memory; and

A processor coupled to the memory, the processor configured to perform the information recommendation method of any one of claims 1 to 14 based on instructions stored in the memory.
A computer-storable medium having computer program instructions stored thereon, the instructions, when executed by a processor, implement the information recommendation method as claimed in any one of claims 1 to 14.