Material recommendation method and device, computer equipment and storage medium

Abstract

The invention relates to a material recommendation method and device, computer equipment and a storage medium, and relates to the technical field of material recommendation. The method comprises the steps that recall scores of at least two materials to be subjected to cold start recall through a cold start recall path are acquired; based on the recall scores of the at least two to-be-cold-started materials, material warranty based on the first number is carried out, and a target material set is obtained; target materials are determined based on the cold start weights of all the materials to be subjected to cold start and transported in the target material set; the cold start weight is determined based on the current exposure speed and the expected exposure speed of the material to be subjected to cold start; and inserting the target material into the material recommendation sequence for material recommendation. Through the method, the long tail effect of material exposure can be reduced through material cold start, and the material to be subjected to cold start can be controlled not to be subjected to overexposure or not to be subjected to underexposure while the material cold start can be completed within the specified time.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of material recommendation, and in particular, to a material recommendation method, device, computer equipment and storage medium. Background Art

[0002] In the recommendation scenarios and post-video recommendation scenarios, in order to provide users with real-time personalized recommendation information flow services, the recommendation system architecture currently widely used in the industry is adopted, that is, materials are recalled from the material library, roughly sorted and finely sorted, and then materials are recommended according to rules.

[0003] However, in the process of personalized recommendation, many materials are screened out, especially those newly added to the material library. Since these materials are not selected by the recommendation system, they lack exposure opportunities, which in turn causes the long-tail problem of material exposure, that is, most of the exposure is concentrated on a small number of materials, and a large number of materials have very low exposure. In the long run, the recommendation system will gradually tend to high-temperature materials after learning, resulting in a gradual weakening of the personalization effect. From the perspective of content recommendation, a large amount of exposure will be concentrated on a small number of materials, thus affecting the material recommendation effect of the recommendation system. Summary of the invention

[0004] The embodiments of the present application provide a material recommendation method, device, computer equipment and storage medium, which can reduce the long-tail effect of material exposure through material cold start, and enable the material cold start to be completed within a specified time while controlling the material to be cold started to be neither overexposed nor underexposed. The technical solution is as follows.

[0005] In one aspect, a material recommendation method is provided, the method comprising:

[0006] Obtaining recall scores of at least two to-be-cold-start materials recalled through the cold-start recall path; the recall scores are used to indicate the relevance between the to-be-cold-start materials and the to-be-recommended users and the importance of the to-be-cold-start materials;

[0007] Based on the recall scores of the at least two materials to be cold started, material delivery based on the first quantity is performed to obtain a target material set;

[0008] Determine the target material based on the cold start weight of each of the cold start materials to be transported in the target material set; the cold start weight is determined based on the current exposure speed and the expected exposure speed of the cold start material;

[0009] The target material is inserted into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

[0010] In another aspect, a material recommendation device is provided, the device comprising:

[0011] A score acquisition module, used to acquire the recall scores of at least two materials to be cold-started recalled through the cold-start recall path; the recall scores are used to indicate the relevance between the materials to be cold-started and the users to be recommended and the importance of the materials to be cold-started;

[0012] A material delivery module, configured to deliver materials based on a first quantity based on the recall scores of the at least two materials to be cold started, to obtain a target material set;

[0013] A material determination module, used for determining a target material based on a cold start weight of each of the cold start materials to be transported in the target material set; the cold start weight is determined based on a current exposure speed and an expected exposure speed of the cold start material;

[0014] The material recommendation module is used to insert the target material into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

[0015] In a possible implementation, the score acquisition module includes:

[0016] A vector acquisition submodule, used to acquire the material vector of each to-be-cold-start material contained in the cold-start material library, and the user vector of the to-be-recommended user;

[0017] A similarity score calculation submodule, used to calculate the similarity score corresponding to each of the materials to be cold-started based on the material vector of each of the materials to be cold-started and the user vector of the user to be recommended; the similarity score is used to characterize the matching degree between the materials to be cold-started and the user to be recommended;

[0018] The score determination submodule is used to determine the sum of the similarity score corresponding to each of the materials to be cold started and the cold start weight corresponding to each of the materials to be cold started as the recall score of each of the materials to be cold started.

[0019] In a possible implementation manner, the device further includes:

[0020] A first information acquisition module is used to acquire exposure information of each material to be cold-started contained in the cold-start material library, wherein the exposure information includes an expected exposure amount, a current exposure amount, an exposure time threshold, and an used exposure time;

[0021] A speed calculation module, used for calculating the current exposure speed and the expected exposure speed of each of the materials to be cold-started based on the exposure information of each of the materials to be cold-started;

[0022] The weight determination module is used to determine the cold start weight corresponding to each of the materials to be cold started based on the current exposure speed and the expected exposure speed of each of the materials to be cold started.

[0023] In a possible implementation manner, it is characterized in that the device further includes:

[0024] A recall module is used to recall a second quantity of materials to be cold started in the cold start material library that are ranked high according to the recall score through the cold start recall path, wherein the second quantity is not less than the first quantity.

[0025] In a possible implementation manner, the device further includes:

[0026] A status acquisition module is used to acquire the life cycle status of each material to be cold started contained in the cold start material library; the life cycle status includes: success status, failure status, pause status and cold start status;

[0027] The recall module is used to recall the materials to be cold started whose life cycle status is the cold start status in the cold start material library through the cold start recall path.

[0028] In a possible implementation manner, the device further includes:

[0029] A second information acquisition module is used to obtain the storage time, expected exposure, current exposure and cold start weight of each cold start material contained in the cold start material library;

[0030] The state determination module is used to determine the life cycle state of each material to be cold started based on the storage time, expected exposure, current exposure and cold start weight of each material to be cold started.

[0031] In a possible implementation, the state determination module is used to:

[0032] When the storage time of the target material to be cold-started is less than the storage time threshold and the current exposure reaches the expected exposure, the life cycle state of the target material to be cold-started is determined to be a success state; the target material to be cold-started is any one of the materials to be cold-started in the cold-start material library;

[0033] When the storage time of the target material to be cold-started is greater than the storage time threshold and the current exposure is less than the expected exposure, determining that the life cycle state of the target material to be cold-started is a failure state;

[0034] When the storage time of the target material to be cold started is less than the storage time threshold, the current exposure is less than the expected exposure, and the cold start weight is less than the weight threshold, or when the target material to be cold started enters the cold start material warehouse, it is determined that the life cycle state of the target material to be cold started is a paused state;

[0035] When the warehousing time of the target material to be cold started is less than the warehousing time threshold, the current exposure is less than the expected exposure, and the cold start weight is greater than the weight threshold, it is determined that the life cycle state of the target material to be cold started is a cold start state.

[0036] On the other hand, a computer device is provided, which includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the above-mentioned material recommendation method.

[0037] On the other hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, and the computer program is loaded and executed by a processor to implement the above-mentioned material recommendation method.

[0038] On the other hand, a computer program product is provided, the computer program product comprising at least one computer program, the computer program being loaded and executed by a processor to implement the material recommendation method provided in the above-mentioned various optional implementations.

[0039] The technical solution provided by this application may have the following beneficial effects:

[0040] The material recommendation method provided in the embodiment of the present application obtains the recall scores of multiple materials to be cold started that are recalled through the cold start recall path, and based on the recall scores, at least two materials to be cold started are delivered based on the first quantity to obtain a target material set, and then the target material is determined from the target material set based on the cold start weight of the material to be cold started, and the cold start weight is dynamically adjusted based on the current exposure speed and the expected exposure speed; the target material is inserted into the material recommendation sequence determined based on the traditional recommendation system for material recommendation. Through the above method, the long tail effect of material exposure can be reduced through material cold start, and in the process of material cold start, quantitative delivery is performed in the rough sorting link to ensure the smooth progress of material cold start. In the fine sorting link, since the cold start weight of the cold start material is dynamically adjusted based on the current exposure speed and the expected exposure speed of the material to be cold started, as the cold start weight of each cold start material changes, the determined target material is also changed accordingly, so that the material cold start can be completed within the specified time while controlling the material to be cold started to be neither over-exposed nor under-exposed.

[0041] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0043] Figure 1 A schematic diagram showing a traditional recommendation system architecture shown in an exemplary embodiment of the present application is shown;

[0044] Figure 2 A flow chart of a material recommendation method provided by an exemplary embodiment of the present application is shown;

[0045] Figure 3 A schematic diagram showing a traditional recommendation system architecture provided by an exemplary embodiment of the present application is shown;

[0046] Figure 4 A flow chart of a material recommendation method provided by an exemplary embodiment of the present application is shown;

[0047] Figure 5 A schematic diagram showing the life cycle state switching of a material to be cold started provided by an exemplary embodiment of the present application is shown;

[0048] Figure 6 A schematic diagram of a material recommendation method provided by an exemplary embodiment of the present application is shown;

[0049] Figure 7 A block diagram of a material recommendation device provided by an exemplary embodiment of the present application is shown;

[0050] Figure 8 It is a structural block diagram of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0051] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0052] Material recommendation refers to selecting and presenting the materials that are most likely to arouse user interest or meet user needs from a large number of materials (such as goods, advertisements, content, services, etc.) based on user portraits in specific scenarios through algorithm models or manual rules. Figure 1A schematic diagram of a traditional recommendation system architecture shown in an exemplary embodiment of the present application is shown in FIG. Figure 1 As shown in the figure, the recommendation process of the traditional recommendation system usually includes recall, rough sorting, fine sorting and material recommendation, among which recall refers to quickly screening out the candidate set that the user may be interested in from the huge material library, greatly reducing the scale of data for subsequent processing; rough sorting refers to preliminary screening and sorting in the recalled candidate set, further reducing the number of materials while ensuring the initial accuracy of the sorting; fine sorting refers to the refined sorting of the materials output after rough sorting, using more complex models and more distant features to predict the user's preference for each material as accurately as possible, generate the final recommendation sequence, and recommend materials based on recommendation rules.

[0053] However, in the process of material recommendation by traditional recommendation systems, materials newly added to the material library are not selected by the recommendation system, resulting in a lack of exposure opportunities, which in turn causes a long-tail problem of material exposure; based on this, an embodiment of the present application provides a material recommendation method, which can give materials the opportunity to be exposed within a limited time (this process is called material cold start), and solve problems that may arise during the cold start process, such as overexposure of materials, inability to enter the exposure process, or too long a material cold start time.

[0054] Figure 2 A flow chart of a material recommendation method provided by an exemplary embodiment of the present application is shown. The method can be executed by a material recommendation device, and the material recommendation device can be implemented as a server, such as Figure 2 As shown, the material recommendation method may include the following steps:

[0055] Step 210, obtaining the recall scores of at least two materials to be cold-started that are recalled through the cold-start recall path; the recall scores are used to indicate the relevance between the materials to be cold-started and the users to be recommended and the importance of the materials to be cold-started.

[0056] Among them, the cold start recall path refers to the path used to recall the cold start materials, which is different from the traditional recall path. The corresponding materials are the cold start materials to be added to the recommendation system and lack user behavior data. Since the cold start materials lack the user's direct user interaction history as a basis, the computer device can rely on other information and strategies in the material recall link to evaluate its relevance and importance to the user to be recommended, that is, calculate the recall score; when determining the relevance between the cold start materials and the user to be recommended, the evaluation can be based on the similarity between the contents. Schematically, if the content keywords of the cold start material A to be recommended are ["technology news", "artificial intelligence", "innovative products"], and the historical behavior of the recommended user X shows the preference keywords ["technology", "AI", "innovation"], the content similarity between the two is calculated to determine the relevance between the two. The importance of the cold start material can be characterized by the cold start weight, which can be determined based on the current exposure speed and the expected exposure speed of the cold start material.

[0057] Step 220: Based on the recall scores of at least two materials to be cold started, materials are delivered based on the first quantity to obtain a target material set.

[0058] In the rough sorting stage of the recommendation system, materials returned from different recall paths will be received, and these materials will be comprehensively predicted, scored and sorted using a rough sorting model, and then the top several materials with the highest scores will be selected as the rough sorting results and submitted to the fine sorting stage; however, if the materials to be cold started are uniformly scored and sorted in the rough sorting stage, the materials to be cold started will not be able to pass the rough sorting stage. Therefore, in an embodiment of the present application, when the materials to be cold started reach the rough sorting stage, the computer device will perform a quantitative delivery operation on the materials to be cold started, that is, after recalling at least two materials to be cold started from the cold start material library, a set number (i.e., the first number) of materials to be cold started are screened out from the at least two recalled materials to be cold started to form a target material set; when performing quantitative delivery, the computer device can sort the materials to be cold started based on the recall scores of each material to be cold started to obtain a sorting result; based on the sorting result, the first number of materials to be cold started with the highest recall scores are screened out to form a target material set, that is, in the rough sorting stage, for materials recalled through the cold start recall path, the first number of materials to be cold started are sent to the fine sorting stage based on the recall score.

[0059] Step 230, determining the target material based on the cold start weight of each of the transported cold start materials in the target material set; the cold start weight is determined based on the current exposure speed and the expected exposure speed of the cold start material.

[0060] In the fine ranking stage, the fine ranking model will be used to predict, score and sort the materials in the rough ranking results submitted in the rough ranking stage, and then the top several materials with the highest scores will be selected to form the material recommendation sequence; similarly, the materials to be cold started cannot pass the fine ranking stage. Therefore, in an embodiment of the present application, the computer device can further screen the materials to be cold started that are escorted from the target material set based on the cold start weight, and select one or several of them as target materials; in a possible implementation method, in order to prevent the insertion of too many materials to be cold started, which will cause the scale data and efficiency data of the recommendation system to decrease too much, the number of target materials can be set to 1 in the computer device, that is, 1 material to be cold started is selected from the target material set as the target material.

[0061] The process of determining the cold start weight of the material to be cold started based on the current exposure speed and the expected exposure speed of the material to be cold started can be carried out in an offline process; wherein, the exposure speed can be calculated from the corresponding exposure amount and exposure time. Therefore, the process of screening out the target material based on the cold start weight is the process of screening out the target material based on the exposure amount and exposure time of the cold start material. By dynamically adjusting the cold start weight, it can be ensured that the corresponding material to be cold started receives a reasonable exposure allocation, thereby achieving effective control of the exposure speed, so as to achieve the purpose of achieving the expected exposure amount within the specified time. At the same time, it can also avoid waste of resources or user fatigue caused by overexposure, and failure to fully expose the material due to underexposure.

[0062] Step 240, inserting the target material into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

[0063] Figure 3 A schematic diagram of a traditional recommendation system architecture provided by an exemplary embodiment of the present application is shown as follows: Figure 3 As shown, in Figure 1 Based on the traditional recommendation system architecture shown in the figure, a cold start recall branch for materials to be cold started is added. Figure 1 After the traditional recommendation architecture shown performs material recall, rough sorting and fine sorting, a material recommendation sequence 310 is obtained; the materials to be cold started that are recalled based on the cold start recall branch are quantitatively delivered in the rough sorting stage, and the target material 320 is obtained after material screening based on the cold start weight in the fine sorting stage, and the target material 320 is inserted into the material recommendation sequence to recommend the materials in the sequence according to the recommendation order indicated by the material recommendation sequence; in a possible implementation method, the computer device can insert the target material into a specified position in the material recommendation sequence; the specified position can be set by relevant personnel based on actual needs or historical recommendation information analysis, and the present application does not limit this.

[0064] In summary, the material recommendation method provided in the embodiment of the present application obtains the recall scores of multiple materials to be cold started that are recalled through the cold start recall path, and based on the recall scores, at least two materials to be cold started are delivered based on the first quantity to obtain a target material set, and then the target material is determined from the target material set based on the cold start weight of the material to be cold started, and the cold start weight is dynamically adjusted based on the current exposure speed and the expected exposure speed; the target material is inserted into the material recommendation sequence determined based on the traditional recommendation system for material recommendation. Through the above method, the long tail effect of material exposure can be reduced through material cold start, and in the process of material cold start, quantitative delivery is performed in the rough sorting link to ensure the smooth progress of material cold start. In the fine sorting link, since the cold start weight of the cold start material is dynamically adjusted based on the current exposure speed and the expected exposure speed of the material to be cold started, as the cold start weight of each cold start material changes, the determined target material is also changed accordingly, so that the material cold start can be completed within the specified time while controlling the material to be cold started to be neither over-exposed nor under-exposed.

[0065] In a possible implementation, when the material recommendation device recalls the materials in the cold start material library, on the one hand, the material recall can be performed based on the life cycle status of the materials to be cold started, on the other hand, the material recall can also be performed based on the recall score of the materials to be cold started, or the two recall methods can be combined to perform the material recall; Figure 4 A flow chart of a material recommendation method provided by an exemplary embodiment of the present application is shown. The method can be executed by a material recommendation device, and the material recommendation device can be implemented as a server, such as Figure 4 As shown, the material recommendation method may include the following steps:

[0066] Step 410, obtaining the life cycle status of each material to be cold started contained in the cold start material library; the life cycle status includes: success status, failure status, pause status and cold start status.

[0067] The cold start material library may be a database for storing materials to be cold started, and the cold start material library may be one of multiple material libraries included in the recommendation system; in the embodiment of the present application, the life cycle status of the materials to be cold started may include a success status, a failure status, a pause status, and a cold start status. The life cycle status of the materials to be cold started may be obtained by a computer device through an offline process and written into the cold start material library corresponding to each material to be cold started.

[0068] The process of determining the life cycle status of each material to be cold started by the computer device can be implemented as follows:

[0069] Obtain the storage time, expected exposure, current exposure and cold start weight of each cold start material in the cold start material library;

[0070] The life cycle status of each material to be cold started is determined based on the storage time, expected exposure, current exposure and cold start weight of each material to be cold started.

[0071] The cold start weight of the material to be cold started may be determined based on the exposure information and exposure time information of the material to be cold started. The process may be implemented as follows:

[0072] Acquire exposure information of each material to be cold started contained in the cold start material library, where the exposure information includes expected exposure, current exposure, exposure time threshold, and used exposure time;

[0073] Based on the exposure information of each material to be cold-started, calculate the current exposure speed and the expected exposure speed of each material to be cold-started;

[0074] Based on the current exposure speed and the expected exposure speed of each material to be cold started, the cold start weight corresponding to each material to be cold started is determined.

[0075] The expected exposure and exposure time thresholds can be set based on actual demand when the cold start material is added to the cold start material library, and the current exposure and used exposure time are obtained by real-time statistics of the system. The current exposure speed is the ratio of the current exposure to the used exposure time; the expected exposure speed can also be called the planned guaranteed speed, which is the ratio of the current remaining exposure to the current remaining exposure time.

[0076] In a possible implementation, a weight calculation formula may be pre-set in the material recommendation device, and the weight calculation formula is used to indicate the relationship between the current exposure speed and the expected exposure speed and the cold start weight. Schematically, the weight calculation formula may be expressed as:

[0077]

[0078] Among them, pid represents the cold start weight of the material to be cold started, initpid represents the initialization pid, which can be set based on actual needs, such as 10; e is the scaling factor, which can also be set based on actual needs, such as 10; expo represents the current exposure, spanTime represents the exposure time used, and the ratio of the two is the current exposure speed, k is the acceleration weight, when spanTime is close to the storage time threshold but the material has not completed the cold start, the computer device can adjust pid by adjusting the k value. expectexpo represents the expected exposure, Indicates the expected exposure speed. The difference between the expected exposure speed and the current exposure speed is the basic value of pid. The slower the exposure speed, the larger the pid will be; when the exposure speed is too fast, the pid will become smaller.

[0079] By bringing the current exposure speed and the expected exposure speed of each material to be cold started into the above weight calculation formula, the cold start weight corresponding to each material to be cold started can be determined.

[0080] Among them, corresponding to the life cycle state of the material to be cold started, when the life cycle state of the material to be cold started is the cold start state, the recommended state of the material to be cold started is set to the privileged state, that is, priority recommendation; when the life cycle state of the material to be cold started is the paused state, the recommended state of the material to be cold started is set to the privileged state, that is, no priority recommendation. Based on this, when recalling materials, the computer device can prioritize the recall of materials to be cold started whose recommended state is the privileged state, that is, whose life cycle state is the cold start state.

[0081] Wherein, when the storage time of the target material to be cold-started is less than the storage time threshold and the current exposure reaches the expected exposure, the life cycle state of the target material to be cold-started is determined to be a success state; the target material to be cold-started is any one of the materials to be cold-started in the cold-start material library;

[0082] When the storage time of the target material to be cold-started is greater than the storage time threshold and the current exposure is less than the expected exposure, the life cycle state of the target material to be cold-started is determined to be a failure state;

[0083] When the storage time of the target material to be cold started is less than the storage time threshold, the current exposure is less than the expected exposure, and the cold start weight is less than the weight threshold, or when the target material to be cold started enters the cold start material warehouse, it is determined that the life cycle state of the target material to be cold started is a suspended state;

[0084] When the warehousing time of the target material to be cold started is less than the warehousing time threshold, the current exposure is less than the expected exposure, and the cold start weight is greater than the weight threshold, the life cycle state of the target material to be cold started is determined to be the cold start state.

[0085] Based on the changes in the exposure amount, exposure speed and storage time of the materials to be cold started, the life cycle status of the materials to be cold started will also change accordingly. Schematically, Figure 5 A schematic diagram of the life cycle state switching of a material to be cold started provided by an exemplary embodiment of the present application is shown. Figure 5As shown, taking the warehousing time threshold of 24 hours as an example, if the exposure of the material to be cold-started reaches the expected exposure within 24 hours, the corresponding material to be cold-started will be marked as Success; if the exposure of the material to be cold-started is lower than the expected exposure and exceeds 24 hours, the corresponding material to be cold-started will be marked as Failed; if the exposure of the material to be cold-started is lower than the expected exposure and is within 24 hours, if the exposure speed is too fast, resulting in the pid being reduced to less than the weight threshold, the corresponding material to be cold-started will be marked as Pending; if the exposure of the material to be cold-started is lower than the expected exposure and is within 24 hours, and the pid exceeds the weight threshold, the corresponding material to be cold-started will be marked as Boosting, that is, cold start.

[0086] In addition, when the materials to be cold started enter the cold start material warehouse, they need to be initialized, that is, the life cycle state of the materials to be cold started is set to the suspended state, the corresponding recommended state is the pending authorization state, and the storage time of the materials to be cold started is recorded.

[0087] Step 420: recall the cold start materials in the cold start material library whose life cycle status is the cold start status through the cold start recall path.

[0088] That is to say, the computer device recalls the cold start materials in the cold start material library whose recommended status is the elevated status.

[0089] Step 430: recall a second quantity of materials to be cold started that are ranked higher in the cold start material library according to the recall score through the cold start recall path, where the second quantity is not less than the first quantity.

[0090] The process of obtaining the recall scores of at least two materials to be cold started can be implemented as follows:

[0091] Obtaining the material vector of each to-be-cold-start material contained in the cold-start material library and the user vector of the to-be-recommended user;

[0092] Based on the material vector of each material to be cold-started and the user vector of the user to be recommended, the similarity score corresponding to each material to be cold-started is calculated; the similarity score is used to characterize the matching degree between the material to be cold-started and the user to be recommended;

[0093] The sum of the similarity score corresponding to each material to be cold started and the cold start weight corresponding to each material to be cold started is determined as the recall score of each material to be cold started.

[0094] Taking the process of obtaining the recall score of any one of at least two materials to be cold-started recalled through the cold-start recall path as an example, the computer device can obtain the material characteristics of the target material to be cold-started, and perform vectorization processing to obtain the material vector of the target material to be cold-started, wherein the material characteristics may include but are not limited to content keywords, material creator information, material attributes (such as format and type, language and region, etc.); obtain the user characteristics of the user to be recommended, and perform vectorization processing to obtain the user vector of the user to be recommended; wherein the user characteristics may include but are not limited to user identification, user interest preference information, user social information and behavior history information, etc.

[0095] After obtaining the material vector of the target material to be cold-started and the user vector of the user to be recommended, an inner product operation is performed on the material vector and the user vector, and the inner product result is determined as the similarity score between the target material to be cold-started and the user to be recommended. The similarity score is used to characterize the degree of match between the material to be cold-started and the user to be recommended. The higher the similarity score, the more matched the material to be cold-started and the user to be recommended, and the lower the similarity score, the less matched the material to be cold-started and the user to be recommended.

[0096] The calculation process of the cold start weight of the target cold start material can refer to the cold start weight calculation process in step 410, which will not be repeated here; after obtaining the similarity score corresponding to the target cold start material and the cold start weight of the target cold start material, the sum of the two is determined as the recall score of the target cold start material.

[0097] In the embodiment of the present application, the recall score of the cold-start material can be used as the basis for recalling the cold-start material in the recall link, and can also be used as the basis for guaranteeing the delivery of the cold-start material in the rough sorting link.

[0098] When the material recommendation device recalls materials based on the material recall scores of the materials to be cold started, the quantity of the recalled materials to be cold started (i.e., the second quantity) may be the same as the quantity of the materials to be cold started that are delivered (i.e., the first quantity); or, the quantity of the recalled materials to be cold started (i.e., the second quantity) may be greater than the quantity of the materials to be cold started that are delivered (i.e., the first quantity).

[0099] In one possible implementation, the material recommendation device may select one of the recall methods to recall the materials to be cold-started; or, in another possible implementation, the material recommendation device may recall part of the materials based on the life cycle status of the materials to be cold-started, and recall another part of the materials based on the recall score of the materials to be cold-started, the two parts of the materials are not repeated, and the total quantity of the two parts of the materials is not less than the first quantity; or, in another possible implementation, the material recommendation device may determine the materials to be cold-started in a cold-start state based on the life cycle status, and further sort the materials to be cold-started in a cold-start state based on the recall score, so as to recall the first second quantity of materials to be cold-started in a cold-start state that are ranked higher according to the recall score; it should be noted that the application of the above two material recall methods can be determined based on actual needs, and the present application does not limit this.

[0100] Step 440: Based on the recall scores of at least two materials to be cold started, materials are delivered based on the first quantity to obtain a target material set.

[0101] Since the recall score can indicate the correlation between the material to be cold-started and the user to be recommended and the importance of the material to be cold-started, the cold-start priority represented by the cold-start weight can be considered when recalling the material to be cold-started, and the user's interest can also be taken into account, so that the behavioral information (such as clicks, playback time, interaction, etc.) subsequently obtained from the cold-start materials in the target material set has a higher relative confidence level.

[0102] Step 450: determine the target material based on the cold start weight of each of the cold start materials to be delivered in the target material set.

[0103] When determining the target material, the computer device can sort the materials to be cold started in the target material set based on the cold start weights of the materials to be cold started in the target material set, and determine the one or several with the highest ranking as the target material, where the quantity of the target material can be set based on actual needs.

[0104] Step 460, inserting the target material into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

[0105] In summary, the material recommendation method provided in the embodiment of the present application obtains the recall scores of multiple materials to be cold started that are recalled through the cold start recall path, and based on the recall scores, at least two materials to be cold started are delivered based on the first quantity to obtain a target material set, and then the target material is determined from the target material set based on the cold start weight of the material to be cold started, and the cold start weight is dynamically adjusted based on the current exposure speed and the expected exposure speed; the target material is inserted into the material recommendation sequence determined based on the traditional recommendation system for material recommendation. Through the above method, the long tail effect of material exposure can be reduced through material cold start, and in the process of material cold start, quantitative delivery is performed in the rough sorting link to ensure the smooth progress of material cold start. In the fine sorting link, since the cold start weight of the cold start material is dynamically adjusted based on the current exposure speed and the expected exposure speed of the material to be cold started, as the cold start weight of each cold start material changes, the determined target material is also changed accordingly, so that the material cold start can be completed within the specified time while controlling the material to be cold started to be neither over-exposed nor under-exposed.

[0106] The material recommendation method provided in the embodiment of the present application includes an online process and an offline process. Taking the material recall based on the life cycle status of the material to be cold started as an example, Figure 6 A schematic diagram of a material recommendation method provided by an exemplary embodiment of the present application is shown. The method can be executed by a material recommendation device, and the material recommendation device can be implemented as a server, such as Figure 6 As shown, the material recommendation method includes an online process and an offline process;

[0107] The offline process is used to calculate the cold start weight of the material to be cold started and determine the life cycle state / priority status of the material to be cold started. The process may include the following steps:

[0108] S601, obtaining material information of the material to be cold started from the cold start material library.

[0109] In this process, the computer device can process the materials to be cold-started one by one, or can process multiple materials to be cold-started simultaneously. This application takes the processing process of one material to be cold-started as an example to illustrate.

[0110] S602, determine whether the material to be cold-started is a new material, if so, execute S603, otherwise, execute S604.

[0111] S603, determine whether there is an expected exposure amount, if so, perform a storage operation, otherwise, perform a queue storage operation.

[0112] Among them, the warehousing operation refers to the operation of associating the expected exposure amount with the corresponding new material and storing it in the cold start material library; queuing warehousing refers to adding the material to be cold started into the queuing warehousing queue, so that it can be associated and stored in the cold start material library after the expected exposure amount is added.

[0113] S604, determining whether the recommended state is an elevated privilege state or a pending elevated privilege state, if so, executing S605, otherwise, associating the pending cold start material with the corresponding life cycle state and storing it in the cold start material library.

[0114] Among them, when the recommended status of the material to be cold started is not the privilege escalation status or the privilege escalation status, the life cycle status of the material to be cold started is a success status or a failure status, which can be determined based on the relationship between the warehousing time and the warehousing time threshold and the relationship between the current exposure and the expected exposure, which will not be repeated here.

[0115] S605, calculating the cold start weight of the material to be cold started.

[0116] The process of calculating the cold start weight can be referred to as follows Figure 4 The relevant contents of the illustrated embodiment will not be repeated here.

[0117] S606, determining whether the cold start weight reaches a weight threshold; if so, updating the recommended status of the material to be cold started to an enhanced status; otherwise, updating the recommended status of the material to be cold started to a pending enhanced status.

[0118] The materials to be cold started are associated with the corresponding life cycle status and recommended status and updated in the cold start material library, so that materials can be recalled based on the recommended status of each material to be cold started in the online process.

[0119] The online process includes the following steps:

[0120] S607, recalling the cold start materials whose recommended status is the privileged status from the cold start material library.

[0121] That is, the materials to be cold started whose life cycle status is the waiting cold start status are recalled from the cold start material library.

[0122] S608, obtaining recall scores of at least two to-be-cold-started materials recalled through the cold-start recall path.

[0123] S609, in the rough sorting stage, based on the recall scores of at least two materials to be cold-started, materials are delivered based on the first quantity to obtain a target material set.

[0124] S610, in the fine sorting stage, the target material is determined based on the cold start weight of each of the cold start materials to be delivered in the target material set.

[0125] S611, inserting the target material into the material recommendation sequence for material recommendation.

[0126] The recommended material sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

[0127] Through the above-mentioned offline process and online process, the material can be cold-started within the specified time, and the material to be cold-started can be controlled to be neither over-exposed nor under-exposed.

[0128] Figure 7 A block diagram of a material recommendation device provided by an exemplary embodiment of the present application is shown. The device can implement the following Figure 2 , Figure 4 or Figure 6 All or part of the steps described in any embodiment, such as Figure 7 As shown, the device comprises:

[0129] The score acquisition module 710 is used to acquire the recall scores of at least two materials to be cold-started that are recalled through the cold-start recall path; the recall scores are used to indicate the relevance between the materials to be cold-started and the users to be recommended and the importance of the materials to be cold-started;

[0130] A material delivery module 720 is used to deliver materials based on a first quantity based on the recall scores of the at least two materials to be cold started, to obtain a target material set;

[0131] A material determination module 730 is used to determine a target material based on the cold start weight of each of the cold start materials to be transported in the target material set; the cold start weight is determined based on the current exposure speed and the expected exposure speed of the cold start material to be transported;

[0132] The material recommendation module 740 is used to insert the target material into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

[0133] In a possible implementation, the score acquisition module 710 includes:

[0134] A vector acquisition submodule, used to acquire the material vector of each to-be-cold-start material contained in the cold-start material library, and the user vector of the to-be-recommended user;

[0135] A similarity score calculation submodule, used to calculate the similarity score corresponding to each of the materials to be cold-started based on the material vector of each of the materials to be cold-started and the user vector of the user to be recommended; the similarity score is used to characterize the matching degree between the materials to be cold-started and the user to be recommended;

[0136] The score determination submodule is used to determine the sum of the similarity score corresponding to each of the materials to be cold started and the cold start weight corresponding to each of the materials to be cold started as the recall score of each of the materials to be cold started.

[0137] In a possible implementation manner, the device further includes:

[0138] A first information acquisition module is used to acquire exposure information of each material to be cold-started contained in the cold-start material library, wherein the exposure information includes an expected exposure amount, a current exposure amount, an exposure time threshold, and an used exposure time;

[0139] A speed calculation module, used for calculating the current exposure speed and the expected exposure speed of each of the materials to be cold-started based on the exposure information of each of the materials to be cold-started;

[0140] The weight determination module is used to determine the cold start weight corresponding to each of the materials to be cold started based on the current exposure speed and the expected exposure speed of each of the materials to be cold started.

[0141] In a possible implementation manner, it is characterized in that the device further includes:

[0142] A recall module is used to recall a second quantity of materials to be cold started in the cold start material library that are ranked high according to the recall score through the cold start recall path, wherein the second quantity is not less than the first quantity.

[0143] In a possible implementation manner, the device further includes:

[0144] A status acquisition module is used to acquire the life cycle status of each material to be cold started contained in the cold start material library; the life cycle status includes: success status, failure status, pause status and cold start status;

[0145] The recall module is used to recall the materials to be cold started whose life cycle status is the cold start status in the cold start material library through the cold start recall path.

[0146] In a possible implementation manner, the device further includes:

[0147] A second information acquisition module is used to obtain the storage time, expected exposure, current exposure and cold start weight of each cold start material contained in the cold start material library;

[0148] The state determination module is used to determine the life cycle state of each material to be cold started based on the storage time, expected exposure, current exposure and cold start weight of each material to be cold started.

[0149] In a possible implementation, the state determination module is used to:

[0150] When the storage time of the target material to be cold-started is less than the storage time threshold and the current exposure reaches the expected exposure, the life cycle state of the target material to be cold-started is determined to be a success state; the target material to be cold-started is any one of the materials to be cold-started in the cold-start material library;

[0151] When the storage time of the target material to be cold-started is greater than the storage time threshold and the current exposure is less than the expected exposure, determining that the life cycle state of the target material to be cold-started is a failure state;

[0152] When the storage time of the target material to be cold started is less than the storage time threshold, the current exposure is less than the expected exposure, and the cold start weight is less than the weight threshold, or when the target material to be cold started enters the cold start material warehouse, it is determined that the life cycle state of the target material to be cold started is a paused state;

[0153] When the warehousing time of the target material to be cold started is less than the warehousing time threshold, the current exposure is less than the expected exposure, and the cold start weight is greater than the weight threshold, it is determined that the life cycle state of the target material to be cold started is a cold start state.

[0154] In summary, the material recommendation device provided in the embodiment of the present application obtains the recall scores of multiple materials to be cold started recalled through the cold start recall path, and based on the recall scores, at least two materials to be cold started are delivered based on the first quantity to obtain a target material set, and then the target material is determined from the target material set based on the cold start weight of the material to be cold started, and the cold start weight is dynamically adjusted based on the current exposure speed and the expected exposure speed; the target material is inserted into the material recommendation sequence determined based on the traditional recommendation system for material recommendation. Through the above device, the long tail effect of material exposure can be reduced through material cold start, and in the process of material cold start, quantitative delivery is performed in the rough sorting link to ensure the smooth progress of material cold start. In the fine sorting link, since the cold start weight of the cold start material is dynamically adjusted based on the current exposure speed and the expected exposure speed of the material to be cold started, as the cold start weight of each cold start material changes, the determined target material is also changed accordingly, so that the material cold start can be completed within the specified time while controlling the material to be cold started to be neither over-exposed nor under-exposed.

[0155] Figure 8The structural block diagram of a computer device 800 shown in an exemplary embodiment of the present application is shown. The computer device can be implemented as the material recommendation device in the above scheme of the present application. The computer device 800 includes a central processing unit (CPU) 801, a system memory 804 including a random access memory (RAM) 802 and a read-only memory (ROM) 803, and a system bus 805 connecting the system memory 804 and the central processing unit 801. The computer device 800 also includes a large-capacity storage device 806 for storing an operating system 809, an application program 810 and other program modules 811.

[0156] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, Erasable Programmable Read Only Memory (EPROM), Electronically-Erasable Programmable Read-Only Memory (EEPROM) flash memory or other solid-state storage technology, CD-ROM, Digital Versatile Disc (DVD) or other optical storage, cassettes, tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that the computer storage medium is not limited to the above. The above-mentioned system memory 804 and mass storage device 806 can be collectively referred to as memory.

[0157] According to various embodiments of the present disclosure, the computer device 800 can also be connected to a remote computer on the network through a network such as the Internet. That is, the computer device 800 can be connected to the network 808 through the network interface unit 807 connected to the system bus 805, or the network interface unit 807 can be used to connect to other types of networks or remote computer systems (not shown).

[0158] The memory also includes at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is stored in the memory. The central processing unit 801 implements all or part of the steps in the material recommendation method shown in the above-mentioned embodiments by executing the at least one instruction, at least one program, code set or instruction set.

[0159] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one computer program is stored, and the computer program is loaded and executed by a processor to implement all or part of the steps in the above-mentioned material recommendation method. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0160] In an exemplary embodiment, a computer program product is also provided, the computer program product comprising at least one computer program, the computer program being loaded and executed by a processor Figure 2 , Figure 4 ,or Figure 6 All or part of the steps of the material recommendation method shown in any embodiment.

[0161] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0162] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A material recommendation method, It is characterized in that The method comprises: Obtaining recall scores of at least two to-be-cold-start materials recalled through the cold-start recall path; the recall scores are used to indicate the relevance between the to-be-cold-start materials and the to-be-recommended users and the importance of the to-be-cold-start materials; Based on the recall scores of the at least two materials to be cold started, material delivery based on the first quantity is performed to obtain a target material set; Determine the target material based on the cold start weight of each of the cold start materials to be transported in the target material set; the cold start weight is determined based on the current exposure speed and the expected exposure speed of the cold start material; The target material is inserted into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

2. The method according to claim 1, It is characterized in that The method further comprises: Obtaining a material vector of each to-be-cold-start material contained in a cold-start material library and a user vector of the to-be-recommended user; Based on the material vectors of the materials to be cold-started and the user vectors of the users to be recommended, calculating the similarity scores corresponding to the materials to be cold-started; the similarity scores are used to characterize the matching degree between the materials to be cold-started and the users to be recommended; The sum of the similarity score corresponding to each of the materials to be cold started and the cold start weight corresponding to each of the materials to be cold started is determined as the recall score of each of the materials to be cold started.

3. The method according to claim 2, It is characterized in that The method further comprises: Acquire exposure information of each material to be cold started contained in the cold start material library, wherein the exposure information includes expected exposure, current exposure, exposure time threshold, and used exposure time; Based on the exposure information of each of the materials to be cold-started, calculating the current exposure speed and the expected exposure speed of each of the materials to be cold-started; Based on the current exposure speed and the expected exposure speed of each of the materials to be cold started, the cold start weight corresponding to each of the materials to be cold started is determined.

4. The method according to claim 3, It is characterized in that The method further comprises: A second quantity of materials to be cold started that are ranked high according to the recall score in the cold start material library is recalled through the cold start recall path, and the second quantity is not less than the first quantity.

5. The method according to claim 1, It is characterized in that The method further comprises: Obtaining the life cycle status of each material to be cold started contained in the cold start material library; the life cycle status includes: success status, failure status, pause status and cold start status; The cold start materials to be cold started in the cold start material library whose life cycle status is the cold start status are recalled through the cold start recall path.

6. The method according to claim 5, It is characterized in that The method further comprises: Obtaining the storage time, expected exposure, current exposure and cold start weight of each cold start material contained in the cold start material library; Based on the storage time, expected exposure, current exposure and cold start weight of each of the materials to be cold started, the life cycle status of each of the materials to be cold started is determined.

7. The method according to claim 6, It is characterized in that The determining of the life cycle status of each of the materials to be cold started based on the storage time, expected exposure, current exposure and cold start weight of each of the materials to be cold started includes: When the storage time of the target material to be cold-started is less than the storage time threshold and the current exposure reaches the expected exposure, the life cycle state of the target material to be cold-started is determined to be a success state; the target material to be cold-started is any one of the materials to be cold-started in the cold-start material library; When the storage time of the target material to be cold-started is greater than the storage time threshold and the current exposure is less than the expected exposure, determining that the life cycle state of the target material to be cold-started is a failure state; When the storage time of the target material to be cold started is less than the storage time threshold, the current exposure is less than the expected exposure, and the cold start weight is less than the weight threshold, or when the target material to be cold started enters the cold start material warehouse, it is determined that the life cycle state of the target material to be cold started is a paused state; When the warehousing time of the target material to be cold started is less than the warehousing time threshold, the current exposure is less than the expected exposure, and the cold start weight is greater than the weight threshold, it is determined that the life cycle state of the target material to be cold started is a cold start state.

8. A material recommendation device, It is characterized in that The device comprises: A score acquisition module, used to acquire the recall scores of at least two materials to be cold-started recalled through the cold-start recall path; the recall scores are used to indicate the relevance between the materials to be cold-started and the users to be recommended and the importance of the materials to be cold-started; A material delivery module, configured to deliver materials based on a first quantity based on the recall scores of the at least two materials to be cold started, to obtain a target material set; A material determination module, used for determining a target material based on a cold start weight of each of the cold start materials to be transported in the target material set; the cold start weight is determined based on a current exposure speed and an expected exposure speed of the cold start material; The material recommendation module is used to insert the target material into a material recommendation sequence for material recommendation; the material recommendation sequence is a material sequence determined after rough sorting and fine sorting of materials recalled through other recall paths.

9. A computer device, It is characterized in that The computer device includes a processor and a memory, the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the material recommendation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, It is characterized in that At least one computer program is stored in the computer-readable storage medium, and the computer program is loaded and executed by the processor to implement the material recommendation method according to any one of claims 1 to 7.

Images