Item recommendation method and device

Abstract

The present invention discloses an item recommendation method and device, and relates to the field of computer technology. A specific implementation of the method includes: determining an item candidate set of the current user based on historical behavior data of several users; wherein the item candidate set includes several candidate items; for each candidate item: determining several similar items corresponding to the candidate item; generating a recall item candidate set of the current user based on each candidate item and its corresponding similar items; determining several target items based on the recall item candidate set and several preset items to be recommended; and recommending the several target items to the current user. This implementation can effectively recommend the items to be recommended that the user is interested in to the user, and the recommendation effect is good.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an item recommendation method and device. Background Art

[0002] In the current era of information explosion, there are a huge number of items on the Internet platform. It is necessary to select the recommended items that users are interested in from the many items and recommend the selected recommended items to users. However, the existing recommendation methods often cannot effectively recommend the recommended items that users are interested in to users. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides an item recommendation method and device, which can effectively recommend items to be recommended that the user is interested in to the user.

[0004] In a first aspect, an embodiment of the present invention provides an item recommendation method, comprising:

[0005] Determine a candidate item set for a current user based on historical behavior data of a number of users; wherein the candidate item set includes a number of candidate items;

[0006] For each candidate item: determining a number of similar items corresponding to the candidate item;

[0007] Generating a candidate set of recalled items for the current user according to each of the candidate items and their corresponding similar items;

[0008] Determine a number of target items according to the candidate set of recalled items and a number of preset items to be recommended;

[0009] The plurality of target items are recommended to the current user.

[0010] Optionally,

[0011] Determining the item candidate set of the current user based on the historical behavior data of several users includes:

[0012] Receive a recommendation request from the current user; wherein the recommendation request includes: an identifier of a historically browsed item of the current user;

[0013] The item candidate set is determined according to the historical behavior data and the historical browsed item identifiers.

[0014] Optionally,

[0015] The determining the item candidate set according to the historical behavior data and the historical browsed item identifiers includes:

[0016] Generate a recommended word list according to the historical behavior data; the recommended word list includes: the item identifier targeted by the historical behavior and several corresponding candidate item identifiers;

[0017] Querying the recommended word list to determine a number of candidate item identifiers corresponding to the historically browsed item identifiers;

[0018] The candidate item set is generated according to the plurality of candidate item identifiers.

[0019] Optionally, further comprising:

[0020] determining image vectors for a number of objects;

[0021] The determining of a number of similar items corresponding to the candidate item includes:

[0022] Selecting the image vector of the candidate object from the image vectors of the plurality of objects;

[0023] Calculating a first similarity value between the candidate object and each of the other objects according to the image vector of the candidate object and the image vectors of other objects among the plurality of objects;

[0024] According to the first similarity value, a plurality of similar objects corresponding to the candidate object are determined from the plurality of other objects.

[0025] Optionally,

[0026] The step of determining the image vectors of the plurality of objects comprises:

[0027] Acquiring image information of the plurality of objects;

[0028] Determining image features of each of the objects based on the image information of the plurality of objects;

[0029] An image vector of each of the objects is generated according to the image features of each of the objects.

[0030] Optionally, further comprising:

[0031] determining certain properties of certain items;

[0032] The determining of a number of similar items corresponding to the candidate item includes:

[0033] Selecting several attributes of the candidate item from among several attributes of the several items;

[0034] Calculating a second similarity value between the candidate item and each of the other items according to the attributes of the candidate item and the attributes of other items among the plurality of items;

[0035] According to the second similarity value, a plurality of similar objects corresponding to the selected object are determined from the plurality of other objects.

[0036] Optionally,

[0037] The step of generating a candidate set of recalled items for the current user according to each of the candidate items and their corresponding similar items includes:

[0038] For each candidate item: selecting a preset number of similar items from a number of similar items corresponding to the candidate item;

[0039] The candidate items and similar items corresponding to the candidate items constitute the candidate set of recalled items of the current user.

[0040] Optionally,

[0041] The step of determining a plurality of target items according to the candidate set of recalled items and a plurality of preset items to be recommended includes:

[0042] For each candidate item in the recalled item candidate set: determine whether the candidate item belongs to the item to be recommended, if so, determine the candidate item as the target item; otherwise, determine whether there is a similar item identical to the item to be recommended among the similar items corresponding to the candidate item, if so, determine the similar item identical to the item to be recommended as the target item.

[0043] Optionally,

[0044] The recommending the plurality of target items to the current user comprises:

[0045] Sorting the target items according to a preset sorting model;

[0046] The target items are recommended to the current user according to the ranking results.

[0047] In a second aspect, an embodiment of the present invention provides an item recommendation device, comprising:

[0048] A candidate item determination module is configured to determine an item candidate set for a current user based on historical behavior data of a number of users; wherein the item candidate set includes a number of candidate items;

[0049] A similar item determination module is configured to: for each candidate item: determine a number of similar items corresponding to the candidate item;

[0050] A recalled item generation module, configured to generate a recalled item candidate set of the current user according to each of the candidate items and their corresponding similar items;

[0051] A target item determination module, configured to determine a number of target items according to the candidate set of recalled items and a number of preset items to be recommended;

[0052] The item recommendation module is configured to recommend the plurality of target items to the current user.

[0053] In a third aspect, an embodiment of the present invention provides an electronic device, including:

[0054] one or more processors;

[0055] a storage device for storing one or more programs,

[0056] When the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the above embodiments.

[0057] In a fourth aspect, an embodiment of the present invention provides a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the above embodiments.

[0058] An embodiment of the above invention has the following advantages or beneficial effects: a candidate set of recalled items for the current user is generated based on each candidate item and its corresponding similar item. The candidate set of recalled items includes not only candidate items but also similar items corresponding to the candidate items. The candidate set of items only includes candidate items. The number of items included in the candidate set of recalled items is greater than that in the candidate set of items. Compared with the recommendation method of directly selecting the target item from the candidate set of items, the range of selectable items is wider when selecting the target item from the candidate set of recalled items. Therefore, the items to be recommended that the user is interested in can be recommended to the user more effectively, and the recommendation effect is better.

[0059] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0061] Figure 1 is a flow chart of an item recommendation method provided by an embodiment of the present invention;

[0062] Figure 2 is a flow chart of an item recommendation method provided by another embodiment of the present invention;

[0063] Figure 3 is a flowchart of an item recommendation method provided by another embodiment of the present invention;

[0064] Figure 4 is a flowchart of an item recommendation method provided by another embodiment of the present invention;

[0065] Figure 5 is a flow chart of an item recommendation method provided by another embodiment of the present invention;

[0066] Figure 6 is a schematic diagram of an item recommendation device provided by an embodiment of the present invention;

[0067] Figure 7 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0068] Figure 8 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0070] The recommendation system mainly includes: recommendation module and item mounting module. The recommendation module can be further divided into recall module and click rate prediction module. The recall module is responsible for quickly selecting candidate items that users may be interested in from the full set of items. The click rate prediction module is responsible for scoring and sorting the candidate items. The system will eventually display the items to the user in the order of their scores.

[0071] The item mounting module stores a number of items to be recommended. Regarding how to recommend items that the user is interested in to the user, there are mainly the following two solutions with a wide range of applications.

[0072] The first solution is to set the item mounting module downstream of the recall module. The item mounting module can be used to identify the items in the item candidate set to be recommended, fill in the relevant information of the items to be recommended, and perceive the real-time changes and adjustments of the information of the items to be recommended. This solution can facilitate the price comparison and deduction logic of the downstream system.

[0073] In the first solution, since the recall module and the item mounting module are two independent modules, there are many items in the item candidate set that are not to be recommended. These items will be filtered in the item mounting module, which ultimately results in a small number of items transmitted to the downstream click rate estimation module, reducing the number of valid items and making the click rate estimation model unable to fully exert its scoring performance. At the same time, it will also lead to insufficient bidding between items, resulting in the loss of relevant platform revenue.

[0074] Since the recall module and the item mounting module are set up independently, the recall module can increase the proportion of items to be recommended in the item candidate set through offline perception item sets and online perception item sets. However, item managers sometimes adjust item bids very frequently, and the offline perception item set of the recall module often cannot obtain a complete and accurate set of items to be recommended in a timely manner. The online perception item set essentially puts the item filtering function in front, and can add new items to be recommended at the end, but these newly added items to be recommended may not be what the user is really interested in. Therefore, if the recall strategy cannot be adjusted in coordination, the online perception item strategy will have limited effect. At the same time, the connection to new external services will significantly increase the time consumption and performance pressure of the recall module.

[0075] The second solution: The recall module and the item mounting module are combined. The recall generates a candidate set of items based on the set of items to be recommended, that is, the recall module can directly output the set of items to be recommended. However, not all recommendation requirements in the system need to display the items to be recommended. The simultaneous existence of recommendation requirements for items to be recommended and recommendation requirements for items not to be recommended will make the system too complicated. In addition, coupling the two modules together will greatly increase the maintenance cost.

[0076] The above implementation methods of recommending items of interest to users all have problems such as: the number of valid items in the item candidate set is small, the maintenance cost is too high, etc. Based on the above analysis, an embodiment of the present invention provides an item recommendation method, such as Figure 1 As shown, including:

[0077] Step 101: Determine a candidate item set for a current user based on historical behavior data of a number of users; wherein the candidate item set includes a number of candidate items.

[0078] The historical behavior data of several users refers to the historical behavior data of several users within the statistical time period. The statistical time period can be set according to specific needs, such as the statistical time period can be the last week of the current time, the last month of the current time, or the last quarter of the current time. The historical behavior data refers to the relevant data of multiple behaviors performed by several users. The historical behavior data includes but is not limited to the relevant data such as browsing clicks, adding to cart, placing orders, adding to follow, etc.

[0079] The item candidate set includes several candidate items, which are items that the current user may be interested in. Recommending the candidate items in the item candidate set to the current user will help the current user to browse, click, add to cart, place an order, and follow.

[0080] There are many ways to determine the item candidate set of the current user based on the historical behavior data of several users. For example, based on the historical behavior data of several users, the degree of association of each item can be determined, and then based on the degree of association of each item and the behavior data of the current user, the item candidate set of the current user can be determined. For example, if multiple users buy badminton rackets after buying badminton balls, it can be determined that there is a strong association between badminton rackets and badminton balls. If the current user has bought a badminton racket, badminton balls of different brands can be recommended to the current user as candidate items.

[0081] A recall algorithm can also be applied to generate a recommended word list based on various user behavior data. The recommended word list is used to store items with strong associations. The candidate set of items for the current user can be determined based on the recommended word list and the current user's behavior data. The recall algorithm may include CF (collaborative filtering), SRS (sequential recommendation algorithm), GraphBased (graph recall algorithm), etc.

[0082] It is also possible to determine the correlation between user behaviors based on the historical behavior data of several users, and determine the candidate set of items for the current user based on the correlation between user behaviors and the behavior data of the current user. For example, if multiple users follow an item and all purchase the item when the price of the item drops by more than 10%, it can be determined that there is a strong correlation between the user's behavior of following and the user's behavior of purchasing. If the current user has followed an item, the item can be recommended to the current user as a candidate item when the price of the item drops by more than 10%.

[0083] Step 102: For each candidate item: determine a number of similar items corresponding to the candidate item.

[0084] Similar items can be items that are identical or similar to the candidate items in several attributes, such as type, brand, usage, origin, price, etc.

[0085] Step 103: Generate a candidate set of recalled items for the current user based on each candidate item and its corresponding similar items.

[0086] The recalled item candidate set contains each candidate item and its corresponding similar items. The recalled item candidate set has a larger number of items than the item candidate set.

[0087] Step 104: Determine a number of target items based on the recalled item candidate set and a number of preset items to be recommended.

[0088] The items to be recommended are items that the system needs to recommend and display to users. The items to be recommended can be set according to specific needs. For example, the items to be recommended can be promotional items in the item promotion system, or advertising items in the advertising mounting system.

[0089] Step 105: Recommend several target items to the current user.

[0090] The target item is an item to be recommended that the system infers that the user may be interested in. The target item may be a subset of the set of items to be recommended in step 104 .

[0091] In the solution of the embodiment of the present invention, the candidate set of recalled items includes not only the candidate items but also the similar items corresponding to the candidate items. The candidate set of items only includes the candidate items. The number of items included in the candidate set of recalled items is greater than that in the candidate set of items. When selecting a target item from the candidate set of recalled items, the range of target items that can be selected is wider and the room for selecting the target item is greater. Therefore, the problem in the prior art that the items to be recommended that the user is interested in cannot be effectively recommended to the user can be solved.

[0092] In addition, the above steps 101 to 103 can be performed by the recall module, and the above step 104 can be performed by the item mounting module. Since the recall module and the item mounting module are independently configured, the method of the embodiment of the present invention can also solve the problem that the recommendation requirements of non-recommended items cannot be met due to coupling the two modules together and the maintenance cost is high.

[0093] In one embodiment of the present invention, generating a candidate set of recalled items for the current user based on each candidate item and its corresponding similar items includes:

[0094] For each candidate item: selecting a preset number of similar items from a number of similar items corresponding to the candidate item;

[0095] Among them, each candidate item and the similar items corresponding to the candidate items constitute the candidate set of recalled items of the current user.

[0096] The preset number can be set according to specific needs and system load conditions. The preset number can be 1, 3, etc.

[0097] Some candidate items may correspond to a large number of similar items. If all similar items corresponding to each candidate item are added to the candidate set of recalled items, the number of items in the candidate set of recalled items may be too large, so that the subsequent process of determining the target item takes too much time.

[0098] In one embodiment of the present invention, a plurality of target items are determined according to the candidate set of recalled items and a plurality of preset items to be recommended, including:

[0099] For each candidate item in the candidate set of recalled items: determine whether the candidate item belongs to the item to be recommended. If so, determine the candidate item as the target item; otherwise, determine whether there is a similar item identical to the item to be recommended among the similar items corresponding to the candidate item. If so, determine the similar item identical to the item to be recommended as the target item.

[0100] Candidate items are determined based on the historical behavior data of several users. Similar items are determined based on the similarity between other items and candidate items. Therefore, the current user's interest in candidate items is usually higher than that in similar items.

[0101] If the candidate item itself is an item to be recommended, the candidate item is determined as the target item. If the candidate item is not an item to be recommended, the target item is selected from similar items. The method of the embodiment of the present invention can make the determined target item more easily noticed by customers and achieve better recommendation effect.

[0102] In one embodiment of the present invention, several target items are recommended to the current user, including:

[0103] Sort a number of target items according to a preset sorting model;

[0104] Recommend several target items to the current user according to the sorting results.

[0105] The sorting model may be a CTR (Click-Through-Rate) sorting model, such as a DCN model or a DeepFM model, etc. Recommending several target items to the current user according to the sorting results can make the display effect of the target items better and the click-through conversion rate of the system higher.

[0106] The present invention provides an item recommendation method. Figure 2 As shown, including:

[0107] Step 201: receiving a recommendation request from the current user; wherein the recommendation request includes: identifiers of items browsed historically by the current user.

[0108] When the current user opens a specific application on the website or clicks a specific button, the system will receive a recommendation request. The historical browsed item ID is the ID of the item browsed by the current user during the statistical time period. The statistical time period can be set according to specific needs, such as the statistical time period can be the last week of the current time, the last month of the current time, or the last quarter of the current time. The item ID can uniquely identify the item in the system.

[0109] Step 202: Determine a candidate set of items based on historical behavior data and historically browsed item identifiers.

[0110] Generate a recommendation word list based on historical behavior data; the recommendation word list includes: the item identifier targeted by the historical behavior and several corresponding candidate item identifiers; query the recommendation word list to determine several candidate item identifiers corresponding to the historical browsed item identifier; generate an item candidate set based on the several candidate item identifiers.

[0111] The recommended word list can be generated by applying the recall algorithm. The recommended word list is used to store items with strong associations. The recommended word list is in key-value sequence format, and each line format is "item 1: item 2, item 3...", the first item 1 is the key, and the subsequent long zipper "item 2, item 3,..." is the value.

[0112] The item candidate set of the current user can be determined based on the recommended word list and the item ID of the current user's historical browsing. Specifically, the value corresponding to the item ID in the recommended word list is queried with the item ID of the current user's historical browsing as the key, and the value is determined as the candidate item.

[0113] Step 203: For each candidate item: determine a number of similar items corresponding to the candidate item.

[0114] Step 204: Generate a candidate set of recalled items for the current user based on each candidate item and its corresponding similar items.

[0115] Step 205: Determine a number of target items based on the recalled item candidate set and a number of preset items to be recommended.

[0116] Step 206: Recommend several target items to the current user.

[0117] In the embodiment of the present invention, a recommendation word list is generated based on the historical behaviors of several users, and then the item candidate set of the current user is determined based on the recommendation word list and the historical browsing item identifiers of the current user. Since the recommendation word list is generated based on the historical behaviors of several users, the recommendation word list more accurately records the association relationship between the items. Therefore, the determined item candidate set of the current user can be more consistent with the user's interests.

[0118] The present invention provides an item recommendation method. Figure 3 As shown, including:

[0119] Step 301: Determine a candidate item set for the current user based on historical behavior data of a number of users; wherein the candidate item set includes a number of candidate items.

[0120] Step 302: Determine image vectors of a number of objects.

[0121] Image vectors are a set of vectors that can be used to characterize the features of item images. Different item images correspond to different item images. In order to achieve better recommendation results, the main image of the item is selected as the item image of the item. The main image of the item is the display image of the item on the recommendation page or the main display image of the item.

[0122] In one embodiment of the present invention, determining image vectors of a plurality of objects comprises:

[0123] Obtain image information of several objects;

[0124] Determine image features of each object based on image information of a plurality of objects;

[0125] According to the image features of each object, an image vector of each object is generated.

[0126] Image information can be obtained from GIF files, PSD files, and other files corresponding to the image. Then, the image features are determined through the image feature extraction algorithm. The image feature extraction algorithm may include: HOG (histogram of Oriented Gradient), SIFT (Scale-invariant features transform), etc. The various features of the image are combined to generate the image vector of the object.

[0127] Step 303: For each candidate object: select an image vector of the candidate object from image vectors of a number of objects.

[0128] The candidate items are the remaining items after removing the candidate items from the plurality of items.

[0129] Step 304: Calculate a first similarity value between the candidate object and each of the other objects according to the image vector of the candidate object and the image vectors of other objects among the plurality of objects.

[0130] The first similarity value is used to characterize the similarity between the main image of the candidate item and other items. The first similarity value can be represented by structural similarity measurement, cosine similarity, mutual information, etc.

[0131] Step 305: Determine a number of similar items corresponding to the candidate item from a number of other items according to the first similarity value.

[0132] A first threshold corresponding to the first similarity value can be set according to needs. If the first similarity value between the main image of the candidate item and the main image of other items is not less than the first threshold, the other item is determined to be a similar item of the candidate item; otherwise, the other item is determined not to be a similar item of the candidate item.

[0133] Step 306: Generate a candidate set of recalled items for the current user based on each candidate item and its corresponding similar items.

[0134] Step 307: Determine a number of target items based on the recalled item candidate set and a number of preset items to be recommended.

[0135] Step 308: Recommend several target items to the current user.

[0136] On the recommendation page, most of the display area is used to display pictures of items. The pictures of items largely determine whether users click on the items. The embodiments of the present invention determine the similarity of items based on the picture information of the items, and can accurately determine similar items of candidate items in a short time, significantly increase the number of items in the item candidate set at the entrance of the click rate estimation model, and improve the recommendation effect without adjusting the recall strategy.

[0137] The present invention provides an item recommendation method. Figure 4 As shown, including:

[0138] Step 401: Determine a candidate item set for the current user based on historical behavior data of a number of users; wherein the candidate item set includes a number of candidate items.

[0139] Step 402: Determine a number of attributes of a number of items.

[0140] Attributes are used to characterize the characteristics of an item. Attributes may include: type, brand, purpose, origin, price, etc. If two items have multiple attributes that are the same or similar, then the two items are similar items.

[0141] Step 403: For each candidate item: select several attributes of the candidate item from several attributes of several items.

[0142] The candidate items are the remaining items after removing the candidate items from the plurality of items.

[0143] Step 404: Calculate a second similarity value between the candidate item and each of the other items based on the attributes of the candidate item and the attributes of the other items in the plurality of items.

[0144] The second similarity value is used to characterize the similarity between the attributes of the candidate item and other items. The second similarity value can be calculated by a variety of methods. For example, the number of attributes that the candidate item has in common with other items can be used as the second similarity value. Different weights can also be set for different attributes, and the sum of the weights of the attributes that the candidate item has in common with other items can be used as the second similarity value.

[0145] Step 405: Determine a number of similar items corresponding to the candidate item from a number of other items according to the second similarity value.

[0146] A second threshold corresponding to the second similarity value can be set according to needs. If the second similarity value between the main image of the candidate item and other items is not less than the second threshold, the other items are determined to be similar items of the candidate item; otherwise, the other items are determined not to be similar items of the candidate item.

[0147] Step 406: Generate a candidate set of recalled items for the current user based on each candidate item and its corresponding similar items.

[0148] Step 407: Determine a number of target items based on the recalled item candidate set and a number of preset items to be recommended.

[0149] Step 408: Recommend several target items to the current user.

[0150] The embodiment of the present invention determines the similarity between items based on item attributes, and can accurately determine similar items of candidate items in a relatively short period of time, significantly increase the number of items in the item candidate set at the entrance of the click-through rate prediction model, and make the target items recommended to the user meet the user's focus.

[0151] Figure 5 It is a flowchart of an item recommendation method provided by another embodiment of the present invention. Figure 5 The vertical part includes offline steps J1, J2 and online steps J3-J10. Figure 5 The horizontal part includes step J3 corresponding to the request module, steps J1-J2 and J4-J8 corresponding to the recall module, and steps J9-J10 corresponding to the item mounting module.

[0152] Step J1: Apply the recall algorithm to generate a recommended word list available online based on various user behavior data (including but not limited to browsing clicks, adding to cart, placing orders, adding to followers, etc.);

[0153] Step J2: Generate an image vector for each item. Based on the main image information of the items in the item library, use the Encoder-Decoder algorithm to train the image features and obtain the main image vector of each item;

[0154] Step J3: Receive advertising requests. When a user opens a website application, an advertising recommendation request message is generated, which contains information such as the user's historical browsing behavior and user portrait;

[0155] Step J4, based on the user's relevant information, use the recommended word list generated in J1 to recall the candidate set of items that the user is interested in. Taking the ICF (Item based Collaborative Filtering) word list as an example, the user's historical browsing behavior is used as the search key, and the corresponding value sequence is used as the candidate set of items;

[0156] Step J5: Generate item candidate set. Merge the vocabulary of various recall methods in step J4 to obtain the complete item candidate set L1, ..., Ln;

[0157] Step J6: Retrieval algorithm: Based on the candidate set of items obtained in step J5, a retrieval algorithm, such as the Faiss algorithm, is used to find several items with the most similar images for each item in the candidate set;

[0158] Step J7: Generate a candidate set of similar items. Perform a deduplication operation on the result obtained in step J6 to obtain a candidate set of similar items;

[0159] Step J8: From the candidate set of similar items generated in step J7, extract the most similar K items. In the figure, K=2, and the main image of item L1 has the most similar items s1 and s2. The extracted item set is merged with the set in step J5 to generate a candidate set of recalled items;

[0160] Step J9: Set up an item mounting module to sense the operations of relevant managers online in real time, update item materials, item advertising unit index, and item premium calculation in real time;

[0161] Step J10: Use the item mounting operation to perform item information calculation and other operations on the recalled item set generated in step J8. As shown in the figure, if the L1 item itself is an item to be recommended, filter out s1 and s2; if the L2 item is not an item to be recommended, traverse similar candidates s3 and s4 in turn, and if s3 is an item to be recommended, directly filter out s4 without further judging the item to be recommended; if all three candidate sets are not items to be recommended, filter out all three candidate sets;

[0162] The results processed by J10 are fed into the CTR ranking model. The first K items are displayed to the user in the order of their scores, where K is a positive integer.

[0163] In the recommendation information flow scenario, each item is usually displayed to the user with three lines of information, including: picture, title, and price. Among them, the picture occupies more than 70% of the display area, and the picture of the picture greatly dominates whether the user clicks on the item. In view of the situation where the recommendation system and the advertisement mounting system are independent of each other, the embodiment of the present invention proposes an advertisement recommendation method based on the main image information of the item, which can greatly increase the system time consumption, significantly increase the number of items in the item candidate set at the entrance of the click-through rate estimation model, improve the recommendation effect without adjusting the recall strategy, and increase the platform advertising revenue.

[0164] The technical key point of the embodiment of the present invention is to adopt an advertising scenario recommendation algorithm based on the main image information of the item for the case where the recommendation module and the item mounting module are the same and independent of each other, capture the user's focus according to the characteristics of the recommendation position of the information flow, learn the feature vector based on the offline training of the main image of the item, and retrieve "same image" items for each recall candidate set online to increase the number of alternative items in the candidate set of recalled items, while greatly increasing the system time consumption, ensuring the positive recommendation effect and increasing the platform's advertising revenue.

[0165] like Figure 6 As shown, an embodiment of the present invention provides an item recommendation device, comprising:

[0166] The recall module 601 is configured to determine a candidate item set of the current user based on the historical behavior data of the plurality of users; wherein the candidate item set includes a plurality of candidate items; for each candidate item: determine a plurality of similar items corresponding to the candidate item; and generate a candidate item recall set of the current user based on each candidate item and its corresponding similar items;

[0167] The item mounting module 602 is configured to determine a number of target items according to the recalled item candidate set and a number of preset items to be recommended; and recommend the number of target items to the current user.

[0168] In one embodiment of the present invention, the device further comprises:

[0169] The request module 603 is configured to receive a recommendation request from the current user; wherein the recommendation request includes: an identifier of a historically browsed item of the current user;

[0170] The recall module 601 is specifically configured to determine a candidate set of items based on the historical behavior data and historical browsed item identifiers.

[0171] In one embodiment of the present invention, the recall module 601 is specifically configured as follows:

[0172] Generate a recommendation word list based on historical behavior data; the recommendation word list includes: the item identifier targeted by the historical behavior and several corresponding candidate item identifiers; query the recommendation word list to determine several candidate item identifiers corresponding to the historical browsed item identifier; generate an item candidate set based on the several candidate item identifiers.

[0173] In one embodiment of the present invention, the recall module 601 is specifically configured as follows:

[0174] Determine image vectors of several objects; select image vectors of candidate objects from the image vectors of several objects; calculate first similarity values ​​between the candidate object and other objects from the several objects based on the image vectors of the candidate object and the image vectors of other objects from the several objects; and determine several similar objects corresponding to the candidate object from the several other objects based on the first similarity values.

[0175] In one embodiment of the present invention, the recall module 601 is specifically configured as follows:

[0176] Obtain image information of a number of objects; determine image features of each object based on the image information of the number of objects; and generate image vectors of each object based on the image features of each object.

[0177] In one embodiment of the present invention, the recall module 601 is specifically configured as follows:

[0178] Determine several attributes of several items; select several attributes of a candidate item from among the several attributes of the several items; calculate a second similarity value between the candidate item and each other item based on the several attributes of the candidate item and several attributes of other items among the several items; and determine several similar items corresponding to the selected item from among the several other items based on the second similarity values.

[0179] In one embodiment of the present invention, the recall module 601 is specifically configured as follows:

[0180] For each candidate item: a preset number of similar items are selected from a number of similar items corresponding to the candidate item; wherein each candidate item and the similar items corresponding to the candidate item constitute a candidate set of recalled items of the current user.

[0181] In one embodiment of the present invention, the item mounting module 602 is specifically configured as follows:

[0182] For each candidate item in the candidate set of recalled items: determine whether the candidate item belongs to the item to be recommended. If so, determine the candidate item as the target item; otherwise, determine whether there is a similar item identical to the item to be recommended among the similar items corresponding to the candidate item. If so, determine the similar item identical to the item to be recommended as the target item.

[0183] In one embodiment of the present invention, the device further comprises:

[0184] A click rate estimation module 604 is configured to sort the target items according to a preset sorting model;

[0185] Recommend several target items to the current user according to the sorting results.

[0186] An embodiment of the present invention provides an electronic device, including:

[0187] one or more processors;

[0188] a storage device for storing one or more programs,

[0189] When one or more programs are executed by one or more processors, the one or more processors implement the method of any of the above embodiments.

[0190] Figure 7 An exemplary system architecture 700 is shown to which the item recommendation method or item recommendation apparatus according to the embodiments of the present invention can be applied.

[0191] like Figure 7 As shown, system architecture 700 may include terminal devices 701, 702, 703, network 704 and server 705. Network 704 is used to provide a medium for communication links between terminal devices 701, 702, 703 and server 705. Network 704 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0192] The user can use the terminal devices 701, 702, 703 to interact with the server 705 through the network 704 to receive or send messages, etc. Various communication client applications can be installed on the terminal devices 701, 702, 703, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, office applications, etc. (only as examples).

[0193] The terminal devices 701 , 702 , and 703 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0194] Server 705 may be a server that provides various services, such as a backend management server (for example only) that provides support for shopping websites browsed by users using terminal devices 701, 702, and 703. The backend management server may analyze and process received data such as product information query requests, and feed back a set of items to be recommended to the terminal device.

[0195] It should be noted that the item recommendation method provided in the embodiment of the present invention is generally executed by the server 705, and accordingly, the item recommendation device is generally set in the server 705. The server 705 determines a number of target items according to the recalled item candidate set and a number of preset items to be recommended, and sends them to the terminal devices 701, 702, and 703 through the network 704.

[0196] It should be understood that Figure 7 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0197] Reference below Figure 8 , which shows a schematic diagram of the structure of a computer system 800 of a terminal device suitable for implementing an embodiment of the present invention. Figure 8 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0198] like Figure 8 As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the system 800 are also stored. The CPU 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0199] The following components are connected to the I / O interface 805: an input section 809 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed, so that a computer program read therefrom is installed into the storage section 808 as needed.

[0200] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the central processing unit (CPU) 801, the above-mentioned functions defined in the system of the present invention are executed.

[0201] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0202] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0203] The modules involved in the embodiments of the present invention may be implemented in software or hardware. The modules described may also be set in a processor. For example, they may be described as: a processor includes a sending module, an acquisition module, a determination module, and a first processing module. The names of these modules do not, in some cases, constitute limitations on the modules themselves. For example, the sending module may also be described as a "module for sending a picture acquisition request to the connected server."

[0204] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes:

[0205] Determine a candidate item set for a current user based on historical behavior data of a number of users; wherein the candidate item set includes a number of candidate items;

[0206] For each candidate item: determining a number of similar items corresponding to the candidate item;

[0207] Generating a candidate set of recalled items for the current user according to each of the candidate items and their corresponding similar items;

[0208] Determine a number of target items according to the candidate set of recalled items and a number of preset items to be recommended;

[0209] The plurality of target items are recommended to the current user.

[0210] In the technical solution of the embodiment of the present invention, a candidate set of recalled items of the current user is generated according to each candidate item and its corresponding similar item. The candidate set of recalled items includes the candidate items and their corresponding similar items. Compared with the recommendation method in which the candidate set of recalled items only includes the candidate items, the number of items in the candidate set of recalled items is increased. By determining the target item to be recommended to the user through the candidate set of recalled items with a large number of items and the items to be recommended, the items to be recommended that the user is interested in can be effectively recommended to the user, and the recommendation effect is good.

[0211] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An item recommendation method, characterized in that: include: Determine a candidate item set for a current user based on historical behavior data of a number of users; wherein the candidate item set includes a number of candidate items; determining image vectors for a number of objects; For each candidate item: determining a number of similar items corresponding to the candidate item; Generating a candidate set of recalled items for the current user according to each of the candidate items and their corresponding similar items; Determine a number of target items according to the candidate set of recalled items and a number of preset items to be recommended; recommending the plurality of target items to the current user; The determining of a number of similar items corresponding to the candidate item includes: Selecting the image vector of the candidate object from the image vectors of the plurality of objects; Calculating a first similarity value between the candidate object and each of the other objects according to the image vector of the candidate object and the image vectors of other objects among the plurality of objects; Determining, according to the first similarity value, a number of similar objects corresponding to the candidate object from among the other objects; The step of determining the image vectors of the plurality of objects comprises: Acquiring image information of the plurality of objects; Determining image features of each of the objects based on the image information of the plurality of objects; Generating an image vector of each of the objects according to the image features of each of the objects; The step of determining a plurality of target items according to the candidate set of recalled items and a plurality of preset items to be recommended includes: For each candidate item in the recalled item candidate set: determine whether the candidate item belongs to the item to be recommended, if so, determine the candidate item as the target item; otherwise, determine whether there is a similar item identical to the item to be recommended among the similar items corresponding to the candidate item, if so, determine the similar item identical to the item to be recommended as the target item, wherein the item to be recommended is the item recommended for display by the system.

2. The method according to claim 1, characterized in that Determining the item candidate set of the current user based on the historical behavior data of several users includes: Receive a recommendation request from the current user; wherein the recommendation request includes: an identifier of a historically browsed item of the current user; The item candidate set is determined according to the historical behavior data and the historically browsed item identifiers.

3. The method according to claim 2, characterized in that The determining the item candidate set according to the historical behavior data and the historical browsed item identifiers includes: Generate a recommended word list according to the historical behavior data; the recommended word list includes: the item identifier targeted by the historical behavior and several corresponding candidate item identifiers; Querying the recommended word list to determine a number of candidate item identifiers corresponding to the historically browsed item identifiers; The candidate item set is generated according to the plurality of candidate item identifiers.

4. The method according to claim 1, characterized in that: Further including: determining certain properties of certain items; The determining of a number of similar items corresponding to the candidate item includes: Selecting several attributes of the candidate item from among several attributes of the several items; Calculating a second similarity value between the candidate item and each of the other items according to the attributes of the candidate item and the attributes of other items among the plurality of items; According to the second similarity value, a plurality of similar items corresponding to the candidate item are determined from the plurality of other items.

5. The method according to claim 1, characterized in that The step of generating a candidate set of recalled items for the current user according to each of the candidate items and their corresponding similar items includes: For each candidate item: selecting a preset number of similar items from a number of similar items corresponding to the candidate item; The candidate items and similar items corresponding to the candidate items constitute the candidate set of recalled items of the current user.

6. The method according to claim 1, characterized in that The recommending the plurality of target items to the current user comprises: Sorting the target items according to a preset sorting model; The target items are recommended to the current user according to the ranking results.

7. An item recommendation device, characterized in that: include: The recall module is configured to determine a candidate item set of the current user based on the historical behavior data of the plurality of users; wherein the candidate item set includes a plurality of candidate items; for each candidate item: determine a plurality of similar items corresponding to the candidate item; and generate a candidate item recall set of the current user based on each candidate item and its corresponding similar items; An item mounting module is configured to determine a number of target items according to the recalled item candidate set and a number of preset items to be recommended; and recommend the number of target items to the current user; The determining of a number of similar items corresponding to the candidate item includes: Selecting the image vector of the candidate object from the image vectors of the plurality of objects; Calculating a first similarity value between the candidate object and each of the other objects according to the image vector of the candidate object and the image vectors of other objects among the plurality of objects; Determining, according to the first similarity value, a number of similar objects corresponding to the candidate object from among the other objects; The step of determining the image vectors of the plurality of objects comprises: Acquiring image information of the plurality of objects; Determining image features of each of the objects based on the image information of the plurality of objects; Generating an image vector of each of the objects according to the image features of each of the objects; The item mounting module is further configured to: for each candidate item in the recalled item candidate set: determine whether the candidate item belongs to the item to be recommended, and if so, determine the candidate item as the target item; otherwise, determine whether there is a similar item identical to the item to be recommended among the similar items corresponding to the candidate item, and if so, determine the similar item identical to the item to be recommended as the target item, wherein the item to be recommended is an item recommended for display by the system.

8. An electronic device, characterized in that: include: one or more processors; a storage device 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 method according to any one of claims 1 to 6.

9. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images