CN111538901B - Article recommendation method and device, server and storage medium - Google Patents

Abstract

The embodiment of the present invention discloses an item recommendation method, device, server and storage medium. The method includes: screening the preset items in the preset item set according to the preset screening rules to select the target item; if the target The recommended times of the item is less than the first exposure threshold, recommending the target item to the preset user searched according to the preset attribute of the target item; if the recommended times of the target item is greater than the first exposure threshold, according to the The click rate of the target item is sorted and recalled in the preset grouping; if the number of recommendations of the target item is greater than the second exposure threshold, the target item is combined with other preset items in the preset item set And sort and recall according to the preset strategy. The technical solution of this embodiment shortens the iteration cycle of new items, and quickly tests out the effect quality of new items, which can meet the high timeliness requirements for items put forward by the recommendation system of the information industry.

Description

Item recommendation method, device, server and storage medium

Technical Field

Embodiments of the present invention relate to artificial intelligence technology, and more particularly to an item recommendation method, device, server and storage medium.

Background Art

With the rapid development of communication technology and the dramatic increase in data volume, users have placed higher demands on timeliness and personalization in the information industry.

There are two main ways to iterate new items in the prior art: one is to establish a certain association with old items through the content of the item itself, and then expose them based on the association. Then the further exposure of the new item depends on the exposure and click situation of the old item. If the old item is clicked less by the user, then its ranking will be relatively backward, then the exposure cycle of the new item will be further extended, and the number of exposures will also be affected by the recommendation of its related old items. The second is to predict the user's rating of the item through a model, and then perform sorting and recall. But generally speaking, the amount of user behavior data is large, the calculation cycle is long, and there is a certain lag, which prolongs the iteration cycle of the item. Therefore, the iteration cycle of new items is relatively long.

At the same time, since old items are fully exposed, their quality can be determined based on user feedback; however, new items cannot be exposed in a short period of time, and their quality cannot be known. In the prior art, new items are generally at a disadvantage according to the unified sorting and recall method, so the probability of recommending new items with high quality to users is greatly reduced, and new items cannot be exposed.

Summary of the invention

The embodiments of the present invention provide an item recommendation method, device, server and storage medium to shorten the iteration cycle of new items and quickly test the effect quality of new items.

In a first aspect, an embodiment of the present invention provides an item recommendation method, comprising:

Filter the preset items in the preset item set according to the preset filtering rules to select the target item;

If the number of recommendations of the target item is less than a first exposure threshold, recommending the target item to a preset user searched according to a preset attribute of the target item;

If the number of recommendations of the target item is greater than the first exposure threshold, sort and recall the target item in the preset grouping according to the click rate of the target item;

If the number of recommendations of the target item is greater than the second exposure threshold, the target item and other preset items in the preset item set are sorted and recalled according to a preset strategy.

In a second aspect, an embodiment of the present invention further provides an item recommendation device, comprising:

A target item screening module is used to screen preset items in a preset item set according to preset screening rules to select target items;

a target item recommendation module, configured to recommend the target item to a preset user searched according to a preset attribute of the target item if the number of recommendations of the target item is less than a first exposure threshold;

A first sorting and recalling module, configured to sort and recall the target item according to the click rate of the target item in the category to which it belongs if the number of recommendations of the target item is greater than the first exposure threshold;

The second sorting and recalling module is used to sort and recall the target object and other preset objects in the preset object set according to a preset strategy if the number of recommendations of the target object is greater than a second exposure threshold.

In a third aspect, an embodiment of the present invention further provides a server, the server comprising:

one or more processors;

a memory for storing one or more programs,

When the one or more programs are executed by the one or more processors, the one or more processors implement the item recommendation method provided by any embodiment of the present invention.

In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to execute the item recommendation method provided by any embodiment of the present invention.

The embodiment of the present invention recommends by screening target items and selects a corresponding sorting and recall strategy according to the exposure amount, thereby solving the problem that the iteration cycle of new items is relatively long and new items cannot be exposed in a short time, resulting in the inability to know the quality. This achieves the effect of shortening the iteration cycle of new items and quickly testing the quality of new items.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of an item recommendation method in Embodiment 1 of the present invention;

FIG2 is a flow chart of an item recommendation method in Embodiment 2 of the present invention;

FIG3 is a flow chart of an item recommendation method in Embodiment 2 of the present invention;

FIG4 is a schematic diagram of the structure of an item recommendation device in Embodiment 3 of the present invention;

FIG5 is a schematic diagram of the structure of a server in Embodiment 4 of the present invention.

DETAILED DESCRIPTION

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

Embodiment 1

FIG1 is a flow chart of an item recommendation method provided by Embodiment 1 of the present invention. This embodiment is applicable to the case of cold-start recommendation of new items. The method may be executed by an item recommendation device, which may be implemented by hardware and/or software. The method specifically includes the following steps:

Step 110: Filter the preset items in the preset item set according to the preset filtering rule to select the target item;

Among them, a preset item set stored in a database is obtained, and the preset item set stores preset items, which need to be recommended to users through a recommendation engine, and relevant information of the preset items is obtained from the database. It is worth noting that the preset items are screened by a certain dimension to determine whether the preset items are new items, and the dimension information is indispensable. Exemplarily, the dimension is the time difference between the appearance, that is, the difference between the current time and the appearance time of the item. The goal of screening is to find new items from the preset item set. Optionally, the preset items in the preset item set are screened according to the preset screening rules to select the target items, including: determining the preset items whose appearance time difference is less than the first threshold as pre-selected target items; removing the pre-selected target items whose appearance time difference is greater than the second threshold and whose recommendation times exceed the preset exposure threshold; wherein the first threshold is greater than the second threshold; and determining the remaining pre-selected target items as target items. The first threshold is the boundary time for distinguishing new items from old items. For preset items whose appearance time difference is greater than or equal to the first threshold, they are considered to be old items. Exemplarily, the first threshold is generally set to 1 day, which can be set according to the timeliness requirements. In addition, for the pre-selected target items whose release time difference is greater than the second threshold, it is necessary to determine the degree of exposure. If they receive more exposure, that is, the number of recommendations exceeds the preset exposure threshold, they need to be eliminated and not selected as target items for recommendation. Exemplarily, the second threshold can be set to 7200 seconds, and the longest running time of other recommendation modules can generally be selected. The preset exposure threshold is mainly obtained through analysis, and generally the average exposure value of other recommendation logics can be taken. The above preset screening rules filter out the target items, and the remaining preset items are uniformly sorted and recalled using the same preset strategy as the old items.

Step 120: If the number of recommendations of the target item is less than the first exposure threshold, recommend the target item to a preset user searched according to the preset attributes of the target item;

Among them, the first exposure threshold is a trial exposure threshold. For target items that reach the trial exposure threshold, it can be considered that they have been recommended to enough users. For target items that are recommended less than the first exposure threshold, they need to be forcibly recommended to increase exposure. In this way, the quality effect of new items can also be determined based on user feedback. The preset attributes can be information such as the category and/or label of the item. The category and/or label information of the items that the user has recently clicked is obtained, and this part of information is matched with the category and/or label information of the new item to be exposed. The user who clicks the matching item is the preset user. A certain number of preset users are forced to recommend the target item.

Step 130: If the number of recommendations of the target item is greater than the first exposure threshold, sort and recall the target item in the preset grouping according to the click rate of the target item;

Among them, for target items whose recommendation times are greater than the first exposure threshold, the click behaviors of these target items are collected and the click-through rates of the target items are calculated. Since these target items are new items, they are grouped according to the categories and/or tags of the target items. In this way, the target items will be in a preset group. The target items are sorted and recalled from large to small according to the click-through rates of the target items in the same preset group.

Step 140: If the number of recommendations of the target item is greater than the second exposure threshold, the target item and other preset items in the preset item set are sorted and recalled according to a preset strategy.

Among them, for the target items whose recommendation times are greater than the second exposure threshold, it is considered that they have entered the range of old items. For such target items, the same preset strategy as the old items can be used for sorting and recall. Exemplarily, the old items are sorted and recalled using strategies such as collaborative filtering or click-through rate (CTR) sorting, so that the new items and the old items are processed differently and finally unified in the sorting method of recall.

The technical solution of this embodiment, by screening target items for recommendation and selecting corresponding sorting and recall strategies according to the exposure amount, solves the problem that the iteration cycle of new items is relatively long and new items cannot be exposed in a short time, resulting in the inability to know the quality, thereby shortening the iteration cycle of new items and quickly testing the quality of new items.

Embodiment 2

FIG2 is a flow chart of an item recommendation method provided by Embodiment 2 of the present invention. The technical solution of this embodiment is further refined on the basis of the above technical solution. The method specifically includes:

Step 210: Filter the preset items in the preset item set according to the preset filtering rule to select the target item;

计算第一曝光阈值。Step 220: Determine a formula according to the first exposure threshold

A first exposure threshold is calculated.

Among them, all_pv represents the total exposure of preset items every day, ucb_ratio represents the proportion of tentative exposure, item_ucb represents the number of new items every day, and item_ucb_ratio represents the proportion of new items exposed. The proportion of tentative exposure is generally between 5% and 10%, and 5% is selected in this embodiment.

Step 230: If the number of recommendations of the target item is less than the first exposure threshold, obtain the target item of the user's click behavior within a preset time;

Step 240, matching the object item with the target item in terms of preset attributes;

Step 250: Determine the user corresponding to the matched object item as the preset user;

Step 260: Recommend the target item to a preset number of preset users.

Among them, the preset attributes of the items can select categories, match the category of the item that the user recently clicked with the category of the new item, and recommend it to the preset users who have successfully matched, and recommend it to a preset number of preset users in total. According to the user's category preference, new items of the same category are recommended, which makes it easier to quickly test the category effect of the new item.

Optionally, before recommending the target item to a preset number of preset users, the method further includes:

计算预设数量，其中，ucb_max_pv表示第一曝光阈值，user_cnt表示预设用户数量。Determine the formula based on the preset quantity

Calculate the preset number, where ucb_max_pv represents the first exposure threshold, and user_cnt represents the preset number of users.

Step 270: If the number of recommendations of the target item is greater than the first exposure threshold, sort and recall the target item in the preset grouping according to the click rate of the target item;

计算目标物品的点击率，其中，ctr_i表示预设物品i的点击率，pv_i表示预设物品i的曝光量，ck_i表示预设物品i的点击量；Optionally, obtain the click behavior for the target item and determine the formula based on the click rate

Calculate the click rate of the target item, where _ctri represents the click rate of preset item i, _pvi represents the exposure of preset item i, and _cki represents the click volume of preset item i;

In the preset grouping, the target items are sorted and recalled from high to low according to their click rate.

Step 280: If the number of recommendations of the target item is greater than the second exposure threshold, the target item and other preset items in the preset item set are sorted and recalled according to a preset strategy.

FIG3 shows a flow chart of an item recommendation method provided by an embodiment of the present invention, and the flow chart shows a flow chart of a more suitable implementation method of a method for solving the item cold start problem of a recommendation system in the information industry. The steps of the implementation method are as follows:

1. Given a collection of all items, the screening criteria for new items in this example are mainly based on the time dimension, so the item's release time (item_produce_time) must be obtained.

2. Filter out items that meet the new item cold start criteria:

1) If the time difference of the item (time_diff, which is the current time minus the item release time) exceeds a certain threshold (max_time_diff, which is generally set to 1 day and can be set according to timeliness requirements), the item is not a new item, and the conventional sorting and recall method is used, jumping to step 9;

2) If the time difference (time_diff) of the item exceeds a threshold (choice_time_diff = 7200, generally the longest running time of other recommendation modules is selected), and the number of recommendations exceeds the exposure threshold (pv_max), although the item is a new item, it has received enough exposure and can be sorted and recalled with old items in a statistical manner, and jump to step 9;

3) The remaining new items meet the conditions for new item cold start, and proceed to step 3.

3. Obtain the category, label and other information of the new item (information on at least one dimension, try to select the dimension with a higher preparation rate). In this case, categories are used because categories can be used to quickly locate the recommended user group, and in this case, category information is relatively accurate.

4. If the number of recommended items reaches the tentative exposure threshold (ucb_max_pv, where the tentative exposure ratio is generally between 5% and 10%, and 5% is selected in this case), proceed to step 7, otherwise proceed to step 5.

5. Match the category of the item that the user recently clicked with the category of the new item and recommend it to the user, a total of ucb_num users. The system recommends new items of the same category based on the user's category preferences, which makes it easier to quickly test the category effect of the new item.

6. If the recommended number of new items reaches the tentative exposure threshold (ucb_max_pv), jump to step 7, otherwise jump to step 5.

7. Collect the click behavior of new items, calculate the click rate (ctr) of new items, and sort and recall them according to the click rate (from large to small) in the same category.

8. If the recommended quantity of the new item reaches the exposure threshold (pv_max), proceed to step 9, otherwise proceed to step 7.

9. Use the same calculation method as old items for sorting and recall.

Because the new item gets enough exposure and turns into an old item, it no longer has the relevant characteristics of a cold start item, thus completing a product cycle from a cold start item to an old item.

Embodiment 3

FIG4 is a schematic diagram of the structure of an item recommendation device provided in Embodiment 4 of the present invention. As shown in FIG4 , the item recommendation device includes:

The target item screening module 410 is used to screen the preset items in the preset item set according to the preset screening rules to select the target item;

A target item recommendation module 420, configured to recommend the target item to a preset user searched according to a preset attribute of the target item if the number of recommendation times of the target item is less than a first exposure threshold;

A first sorting and recalling module 430 is used to sort and recall the target item according to the click rate of the target item in the category to which it belongs if the number of recommendations of the target item is greater than the first exposure threshold;

The second sorting and recalling module 440 is configured to sort and recall the target object together with other preset objects in the preset object set according to a preset strategy if the number of recommendation times of the target object is greater than the second exposure threshold.

The technical solution of this embodiment, by screening target items for recommendation and selecting corresponding sorting and recall strategies according to the exposure amount, solves the problem that the iteration cycle of new items is relatively long and new items cannot be exposed in a short time, resulting in the inability to know the quality. It achieves the effect of shortening the iteration cycle of new items and quickly testing the quality of new items.

Optionally, the target item screening module 410 includes:

A pre-selected target item determination unit, configured to determine preset items whose release time difference is less than a first threshold as pre-selected target items;

A rejection unit, used to reject pre-selected target items whose release time difference is greater than a second threshold and whose recommendation times exceed a preset exposure threshold; wherein the first threshold is greater than the second threshold;

The target item determination unit is used to determine the remaining pre-selected target items as target items.

Optionally, the target item recommendation module 420 is specifically configured to:

If the number of recommendations of the target item is less than the first exposure threshold, the target item of the user's click behavior within a preset time is obtained;

Matching the object item with the target item in terms of preset attributes;

Determine the user corresponding to the matched object item as the preset user;

Recommend target items to a preset number of preset users.

Optionally, the item recommendation device further includes:

计算预设数量，其中，ucb_max_pv表示第一曝光阈值，user_cnt表示预设用户数量。A preset quantity determination module is used to determine the formula according to the preset quantity before recommending the target item to the preset number of preset users.

Calculate the preset number, where ucb_max_pv represents the first exposure threshold, and user_cnt represents the preset number of users.

Optionally, the item recommendation device further includes:

计算第一曝光阈值，其中，all_pv表示预设物品每天曝光量总和，ucb_ratio表示试探性曝光的占比，item_ucb表示每天新物品的数量，item_ucb_ratio表示曝光新物品的占比。The first exposure threshold determination module is used to determine the formula according to the first exposure threshold before recommending the target item to the preset user searched according to the preset attribute of the target item if the number of recommendation times of the target item is less than the first exposure threshold.

Calculate the first exposure threshold, where all_pv represents the total exposure of preset items every day, ucb_ratio represents the proportion of tentative exposure, item_ucb represents the number of new items every day, and item_ucb_ratio represents the proportion of new items exposed.

Optionally, the first sorting and recalling module 430 is specifically configured to:

计算目标物品的点击率，其中，ctr_i表示预设物品i的点击率，pv_i表示预设物品i的曝光量，ck_i表示预设物品i的点击量；Get the click behavior for the target item and determine the formula based on the click rate

Calculate the click rate of the target item, where _ctri represents the click rate of preset item i, _pvi represents the exposure of preset item i, and _cki represents the click volume of preset item i;

In the preset grouping, the target items are sorted and recalled from high to low according to their click rate.

The item recommendation device provided in the embodiment of the present invention can execute the item recommendation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

Embodiment 4

Figure 5 is a schematic diagram of the structure of a server provided in Embodiment 4 of the present invention. As shown in Figure 5, the server includes a processor 510 and a memory 520. The number of processors 510 in the server may be one or more, and Figure 5 takes one processor 510 as an example. The processor 510 and the memory 520 in the server may be connected via a bus or other means, and Figure 5 takes the connection via a bus as an example.

The memory 520 is a computer-readable storage medium that can be used to store software programs, computer executable programs and modules, such as program instructions/modules corresponding to the item recommendation method in the embodiment of the present invention (for example, the target item screening module 410, the target item recommendation module 420, the first sorting and recalling module 430 and the second sorting and recalling module 440 in the item recommendation device). The processor 510 executes various functional applications and data processing of the server by running the software programs, instructions and modules stored in the memory 520, that is, realizing the above-mentioned item recommendation method.

The memory 520 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 520 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 520 may further include a memory remotely arranged relative to the processor 510, and these remote memories may be connected to the server via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

Embodiment 5

Embodiment 5 of the present invention further provides a storage medium containing computer executable instructions, wherein the computer executable instructions, when executed by a computer processor, are used to perform an item recommendation method, including:

Filter the preset items in the preset item set according to the preset filtering rules to select the target item;

If the number of recommendations of the target item is less than a first exposure threshold, recommending the target item to a preset user searched according to a preset attribute of the target item;

If the number of recommendations of the target item is greater than the first exposure threshold, sort and recall the target item in the preset grouping according to the click rate of the target item;

When the number of recommendations of the target item is greater than a second exposure threshold, the target item and other preset items in the preset item set are sorted and recalled according to a preset strategy.

Of course, the storage medium containing computer executable instructions provided by an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in the item recommendation method provided by any embodiment of the present invention.

Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

It is worth noting that in the embodiment of the above-mentioned item recommendation device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims (10)

1. An item recommendation method, comprising:

screening preset articles in the preset article set according to preset screening rules, and selecting target articles;

if the recommended times of the target object are smaller than a first exposure threshold value, recommending the target object to a preset user searched according to the preset attribute of the target object;

if the recommended times of the target object are greater than the first exposure threshold, sorting recall is performed in a preset group according to the click rate of the target object;

and if the recommended times of the target articles are larger than a second exposure threshold, sorting and recalling the target articles and other preset articles in the preset article set according to a preset strategy.

2. The method of claim 1, wherein the screening the preset items in the set of preset items according to the preset screening rule, and selecting the target item comprises:

determining the preset article with the appearance time difference smaller than a first threshold value as a preselected target article;

removing the preselected target object with the appearance time difference larger than a second threshold and the recommended times exceeding a preset exposure threshold; wherein the first threshold is greater than the second threshold;

and determining the remaining preselected target items as the target items.

3. The method according to claim 1, wherein if the recommended number of times of the target item is smaller than a first exposure threshold, recommending the target item to a preset user searched according to a preset attribute of the target item, comprises:

if the recommended times of the target object are smaller than the first exposure threshold, acquiring the object of the clicking action of the user in the preset time;

matching the object article with the target article in terms of the preset attribute;

determining the matched user corresponding to the object article as the preset user;

recommending the target object to a preset number of preset users.

4. A method according to claim 3, further comprising, prior to said recommending said target item to a predetermined number of said predetermined users:

determining a formula according to a preset number

And calculating the preset number, wherein ucb _max_pv represents the first exposure threshold value, and the user_cnt represents the preset user number.

5. The method of claim 4, further comprising, before recommending the target item to a preset user searched according to the preset attribute of the target item if the recommended number of times of the target item is less than a first exposure threshold value:

determining a formula according to the first exposure threshold

Calculating the first exposure threshold, wherein all_pv represents the sum of the daily exposure amounts of the preset articles, ucb _ratio represents the duty ratio of heuristic exposure, item_ ucb represents the number of new articles per day, and item_ ucb _ratio represents the duty ratio of the new articles to be exposed.

6. The method of claim 1, wherein if the recommended number of times of the target item is greater than the first exposure threshold, sorting recalls in a preset group according to a click rate of the target item, comprising:

acquiring clicking behaviors aiming at the target object, and determining a formula according to the clicking rate

Calculating the click rate of the target object, wherein ctr _i Indicating the click rate of the preset item i, pv _i Indicating the exposure of the preset article i, ck _i Indicating the click rate of a preset article i;

and in the preset group, sorting recalls according to the click rate of the target object from large to small.

7. An article recommendation device, comprising:

the target article screening module is used for screening preset articles in the preset article set according to preset screening rules and selecting target articles;

the target article recommending module is used for recommending the target article to a preset user searched according to the preset attribute of the target article if the recommending times of the target article are smaller than a first exposure threshold value;

the first sorting recall module is used for sorting recall according to the click rate of the target object in the belonging category if the recommended number of times of the target object is larger than the first exposure threshold;

and the second sorting recall module is used for sorting recall of the target object and other preset objects in the preset object set according to a preset strategy if the recommended times of the target object are larger than a second exposure threshold.

8. The apparatus of claim 7, wherein the target item screening module comprises:

a preselected target object determining unit for determining the preset object with the appearance time difference smaller than a first threshold value as a preselected target object;

the rejecting unit is used for rejecting the preselected target object, the appearing time difference of which is larger than a second threshold value, and the recommended times of which exceed a preset exposure threshold value; wherein the first threshold is greater than the second threshold;

and the target object determining unit is used for determining the remaining preselected target objects as the target objects.

9. A server, the server comprising:

one or more processors;

a memory for storing one or more programs;

the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the item recommendation method of any one of claims 1-6.

10. A storage medium containing computer executable instructions which, when executed by a computer processor, are for performing the item recommendation method of any one of claims 1 to 6.

Images