Information processing method and apparatus, storage medium, and electronic device

Abstract

The application discloses an information processing method and device, a storage medium and an electronic equipment. The method comprises the following steps: receiving search information; processing the search information by a target recommendation algorithm to obtain a search result, the target recommendation algorithm being obtained by iteratively training an initial recommendation algorithm by a training sample set constructed by a plurality of target data information, the plurality of target data information being obtained by screening a plurality of historical data information, the historical data information at least comprising a plurality of search result sets and behavior feedback data for search results in the plurality of search result sets, wherein the behavior feedback data is used to represent whether the search result is selected; and displaying the search result. The application solves the technical problem that the accuracy of the search result output by the iteratively updated algorithm model is relatively low after the algorithm model is deployed and the algorithm model is directly iteratively updated according to the feedback data of the user.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to an information processing method and apparatus, a storage medium and an electronic device. Background Technology

[0002] Enterprise-oriented industry solutions incorporate numerous intelligent algorithm models. After deployment to customer sites, these models typically require continuous iteration and upgrades based on changes in customer tasks and data to maintain optimal performance. A common practice is to manually collect user feedback, construct an error sample set, and then retrain the model in the customer's environment. A significant challenge here is that most customers lack GPU resources, and considering cost, the CPU and memory resources available for training are also very limited. Furthermore, mainstream algorithm models now have increasingly more parameters, requiring larger datasets and longer training times. For example, a BERT (Bidirectional Encoder Representation from Transformers) model with 100 million parameters and 1 million training samples can take over ten days to train on a 4-core, 16GB memory machine. Current technology involves periodically collecting a batch of samples after model deployment and then retraining the model on the customer's machine using these samples. However, this approach, by training without sample selection or relying on subjective human selection, results in a massive number of training samples (especially for high-traffic task algorithms) and a significant amount of low-quality data mixed in.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides an information processing method and apparatus, a storage medium and an electronic device to at least solve the technical problem that after deploying an algorithm model, directly updating the algorithm model based on user feedback data results in a relatively low accuracy of the search results output by the iteratively updated algorithm model.

[0005] According to one aspect of the embodiments of this application, an information processing method is provided, comprising: receiving search information; processing the search information through a target recommendation algorithm to obtain search results, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm with a training sample set constructed from multiple target data information, the multiple target data information being obtained by filtering multiple historical data information, the multiple historical data information being filtered according to the target matching degree of the historical data information, the target matching degree representing the degree of matching between the search result set corresponding to the historical data information and the historical search information, the historical data information including at least: multiple search result sets output by the initial recommendation algorithm for the multiple historical search information and behavioral feedback data for the search results in the multiple search result sets, wherein the behavioral feedback data is used to represent whether the search result is selected; and displaying the search results.

[0006] Furthermore, before processing the search information using the target recommendation algorithm to obtain search results, the method further includes: acquiring the plurality of historical data information; filtering the plurality of historical data information to obtain the plurality of target data information, and constructing the training sample set based on the plurality of target data information, wherein the training sample set includes a plurality of training samples, the training samples consisting of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample being the selected search result in the plurality of search results set determined based on the behavioral feedback data, and the first negative sample being the unselected search result in the plurality of search results set determined based on the behavioral feedback data; iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

[0007] Further, the filtering process for the multiple historical data information to obtain multiple target data information includes: for the historical data information in the multiple historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, wherein the target matching degree characterizes the degree of matching between the search result set and the historical search information; filtering the multiple historical data information based on the target matching degree to obtain the multiple target data information.

[0008] Further, for historical data information among multiple historical data information, the target matching degree corresponding to the historical data information is determined based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, using at least one of the following methods: For historical data information among multiple historical data information, based on the behavioral feedback data, it is determined whether the search results in the search result set corresponding to the historical data information are selected; if the search results in the search result set corresponding to the historical data information are selected, the order of the selected search results in the search result set corresponding to the historical data information is determined, and the target matching degree corresponding to the historical data information is determined based on whether the search results are selected and the order of the selected search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, the target matching degree corresponding to the historical data information is determined based on the behavioral feedback data and the target search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, the target matching degree corresponding to the historical data information is obtained based on the category information corresponding to the target search results in the search result set corresponding to the historical data information.

[0009] Furthermore, for historical data information among multiple historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search results in the search results set corresponding to the historical data information includes: for historical data information among multiple historical data information, determining the target search results in the search results set corresponding to the historical data information based on the behavioral feedback data; determining the order of the target search results in the search results set corresponding to the historical data information; calculating the matching degree between the search results set corresponding to the historical data information and the historical search information based on the order of the order, to obtain the target matching degree of the historical data information.

[0010] Furthermore, for historical data information among multiple historical data information, obtaining the target matching degree corresponding to the historical data information based on the category information corresponding to the target search result in the search result set corresponding to the historical data information includes: obtaining the category information corresponding to the search result in the search result set corresponding to the historical data information; calculating the category entropy corresponding to the search result set in the search result set corresponding to the historical data information based on the category information; and determining the target matching degree corresponding to the historical data information based on the category entropy.

[0011] Furthermore, constructing a training sample set based on the multiple target data information includes: filtering the multiple target data information based on the historical search information to obtain filtered target data information; and constructing the training sample set based on the filtered target data information.

[0012] Further, filtering the multiple target data information based on the historical search information to obtain the filtered target data information includes: vectorizing the historical search information to obtain a first vector; vectorizing the target data information in the multiple target data information to obtain multiple second vectors; calculating the cosine similarity between the first vector and the multiple second vectors to obtain multiple cosine similarities; and filtering the multiple target data information based on the multiple cosine similarities to obtain the filtered target data information.

[0013] Further, iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm includes: constructing a sample pool based on historical data information other than the multiple target data information; determining multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool; iteratively training the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0014] Further, determining the multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool includes: calculating the gradient of the sample data in the sample pool according to the initial recommendation algorithm to obtain the gradient vector corresponding to the sample data; clustering according to the gradient vector corresponding to the sample data to obtain multiple initial sample data; and filtering the multiple initial sample data to obtain the multiple second positive samples and the multiple second negative samples.

[0015] According to another aspect of the embodiments of this application, an information processing method is also provided, comprising: receiving search information of a target object uploaded by a client; processing the search information in a cloud server using a target recommendation algorithm to obtain search results, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information; filtering the multiple historical data information based on the target matching degree of the historical data information, wherein the target matching degree characterizes the degree of matching between the search result set corresponding to the historical data information and the historical search information; the multiple target data information is obtained by filtering the multiple historical data information; the historical data information includes at least: a search result set output by the initial recommendation algorithm on the historical search information and behavioral feedback data for the search results in the search result set, wherein the behavioral feedback data is used to characterize whether the search result is selected; and feeding back the search results to the client.

[0016] According to another aspect of the embodiments of this application, an information processing apparatus is also provided, comprising: a receiving unit for receiving search information; a processing unit for processing the search information using a target recommendation algorithm to obtain search results, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information, and filtering the multiple historical data information based on the target matching degree of the historical data information, wherein the target matching degree characterizes the degree of matching between the search result set corresponding to the historical data information and the historical search information, the multiple target data information is obtained by filtering the multiple historical data information, and the historical data information includes at least: multiple search result sets output by the initial recommendation algorithm for the multiple historical search information and behavioral feedback data for the search results in the multiple search result sets, wherein the behavioral feedback data is used to characterize whether the search result is selected; and a display unit for displaying the search results.

[0017] Furthermore, the apparatus further includes: an acquisition unit, configured to acquire the plurality of historical data information before processing the search information through the target recommendation algorithm to obtain search results; a filtering unit, configured to filter the plurality of historical data information to obtain the plurality of target data information, and construct the training sample set based on the plurality of target data information, wherein the training sample set includes a plurality of training samples, the training samples being composed of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample being the selected search result in the plurality of search results set determined based on the behavioral feedback data, and the first negative sample being the unselected search result in the plurality of search results set determined based on the behavioral feedback data; and a training unit, configured to iteratively train the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

[0018] Further, the filtering unit includes: a first determining subunit, used to determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the multiple search result sets, wherein the target matching degree characterizes the degree of matching between the multiple search result sets and the multiple historical search information; and a first filtering subunit, used to filter the multiple historical data information based on the multiple target matching degrees to obtain the multiple target data information.

[0019] Further, the first determining subunit includes at least one of the following: a first determining module, configured to, for historical data information from multiple historical data information, determine whether a search result in the search result set corresponding to the historical data information is selected based on the behavioral feedback data; and if a search result in the search result set corresponding to the historical data information is selected, determine the order of the selected search result in the search result set corresponding to the historical data information, and determine the target matching degree corresponding to the historical data information based on whether the search result is selected and the order of the selected search result in the search result set corresponding to the historical data information; a second determining module, configured to, for historical data information from multiple historical data information, determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search result in the search result set corresponding to the historical data information; and a third determining module, configured to, for historical data information from multiple historical data information, obtain the target matching degree corresponding to the historical data information based on the category information corresponding to the search result in the search result set corresponding to the historical data information.

[0020] Further, the second determining module includes: a first determining submodule, used to determine, based on the behavioral feedback data, the target search result in the search results corresponding to the historical data information from among multiple historical data information; a second determining submodule, used to determine the order of the target search result in the search results set corresponding to the historical data information; and a first calculating submodule, used to calculate the degree of matching between the search results set corresponding to the historical data information and the historical search information based on the order of matching, to obtain the target matching degree of the historical data information.

[0021] Furthermore, the third determining module includes: an acquisition submodule, used to acquire category information corresponding to the search results in the search results set corresponding to the historical data information; a second calculation submodule, used to perform calculations based on the category information to obtain the category entropy corresponding to the search results set corresponding to the historical data information; and a third determining submodule, used to determine the target matching degree corresponding to the historical data information based on the category entropy.

[0022] Furthermore, the filtering unit further includes: a second filtering subunit, used to filter the multiple target data information based on the historical search information to obtain filtered target data information; and a construction subunit, used to construct the training sample set based on the filtered target data information.

[0023] Further, the second filtering subunit includes: a first processing module for vectorizing the historical search information to obtain a first vector; a second processing module for vectorizing the target data information in the plurality of target data information to obtain a plurality of second vectors; a first calculation module for calculating the cosine similarity between the first vector and the plurality of second vectors to obtain a plurality of cosine similarities; and a first filtering module for filtering the plurality of target data information based on the plurality of cosine similarities to obtain filtered target data information.

[0024] Further, the training unit includes: a construction subunit, used to construct a sample pool based on historical data information other than the multiple target data information; a second determination subunit, used to determine multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool; and a training subunit, used to iteratively train the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0025] Further, the second determining subunit includes: a second calculation module, used to calculate the gradient of the sample data in the sample pool according to the initial recommendation algorithm, to obtain the gradient vector corresponding to the sample data; a clustering module, used to cluster according to the gradient vector corresponding to the sample data, to obtain multiple initial sample data; and a second filtering module, used to filter the multiple initial sample data to obtain the multiple second positive samples and the multiple second negative samples.

[0026] In this embodiment, the algorithm receives search information, processes the search information using a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information, which includes at least: multiple search result sets output by the initial recommendation algorithm on multiple historical search information and behavioral feedback data on the search results in the multiple search result sets. The behavioral feedback data is used to characterize whether a search result is selected. The search results are then displayed. This approach solves the technical problem that directly updating the algorithm model using user feedback data after deployment leads to low accuracy of the search results output by the iteratively updated algorithm model. In this solution, filtering historical data information removes low-quality data, resulting in high-quality target data information. Furthermore, filtering historical data information effectively reduces the amount of data used for iterative training, ensuring and improving the effectiveness of iterative training of the initial recommendation algorithm. Iteratively updating the initial recommendation algorithm using high-quality target data information effectively improves the predictive performance of the initial recommendation algorithm, thereby increasing the accuracy of the search results output by the iteratively updated algorithm model. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0028] Figure 1 This is a schematic diagram of a computer terminal provided according to Embodiment 1 of this application;

[0029] Figure 2 This is a flowchart of the information processing method provided in Embodiment 1 of the present invention;

[0030] Figure 3 This is a schematic diagram of the screening of sample data provided in Embodiment 1 of the present invention;

[0031] Figure 4This is a flowchart of the information processing method provided in Embodiment 1 of the present invention. Figure 1 ;

[0032] Figure 5 This is a flowchart of the information processing method provided in Embodiment 1 of the present invention. Figure 2 ;

[0033] Figure 6 This is a flowchart of the information processing method provided in Embodiment 2 of the present invention;

[0034] Figure 7 This is a schematic diagram of an information processing device provided according to Embodiment 3 of the present invention;

[0035] Figure 8 This is a schematic diagram of a computer terminal provided according to Embodiment 4 of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0037] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0038] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0039] Training sample triples: Triples are generally (anchor, positive, negative), where anchor represents a baseline input sample, positive represents a sample that matches the baseline input class or label (positive sample), and negative represents a sample that does not match the baseline input class (negative sample).

[0040] Recall model: The recall model selects a portion of data from a large dataset, sorts it, and finally pushes the sorted results to the target object.

[0041] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, user behavior feedback data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0042] Example 1

[0043] According to an embodiment of this application, an information processing method is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0044] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing an information processing method is shown. Figure 1 As shown, the computer terminal (or mobile device) 10 may include a processor set 102 (the processor set 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA, and the processor set 102 may include a processor set, Figure 1 (Illustrated using 102a, 102b, ..., 102n), a memory 104 for storing data, and a transmission module 106 for communication functions. In addition, it may include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0045] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be wholly or partially embodied in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be wholly or partially integrated into any other element within the computer terminal 10 (or mobile device).

[0046] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the information processing method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the above-mentioned information processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0047] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0048] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0049] Under the aforementioned operating environment, this application provides the following: Figure 2 The information processing method shown. Figure 2 This is a flowchart of an information processing method according to Embodiment 1 of this application. The information processing method described above includes the following steps:

[0050] Step S201: Receive search information.

[0051] Optionally, the search information input by the user (i.e., the target object mentioned above) can be obtained through a target recommendation algorithm. It should be noted that the search information can be searching for some data, or it can be the video or music that the user expects the target recommendation algorithm to recommend.

[0052] Step S202: The search information is processed by a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm with a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information. The multiple historical data information is filtered according to the target matching degree of the historical data information. The target matching degree represents the degree of matching between the search result set corresponding to the historical data information and the historical search information. The historical data information includes at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets. The behavioral feedback data is used to represent whether the search result is selected.

[0053] Optionally, after obtaining the aforementioned search information, the target recommendation algorithm can process the search information and user's historical behavior data to obtain the corresponding search results. It should be noted that the target recommendation algorithm can be a recall algorithm. When making predictions and recommendations, the recall algorithm recommends the top-K results to the user in order. For example, if K=20, the first 20 results predicted by the recall algorithm are recommended to the user in order from 1 to 20. It should be noted that the larger the sequence number, the lower the confidence level and the less reliable the prediction result.

[0054] It's important to note that the target recommendation algorithm is iteratively trained using a training sample set constructed from target data. This target data includes, but is not limited to, multiple search result sets output by the initial recommendation algorithm based on various historical search data, and behavioral feedback data on the search results within these sets. Behavioral feedback data represents user behavior towards the search results output by the initial recommendation algorithm, including but not limited to liking, agreeing, and disagreeing. This behavioral feedback data indicates whether a search result is selected; simply put, it reflects whether the user approves of the search results output by the initial recommendation algorithm. Iterative training using behavioral feedback data can effectively improve the recommendation algorithm's performance.

[0055] Because a large amount of user behavior feedback data is of low quality, directly using this data for iterative training of the recommendation model will not effectively improve the model's recommendation performance and will also lead to excessively long training times. To avoid these problems, the target data information mentioned above is obtained by filtering multiple historical data sets. Filtering historical data effectively improves the quality of the training sample set.

[0056] Step S203: Display the search results.

[0057] Optionally, the search results output by the target recommendation algorithm are pushed to the target object in order.

[0058] In summary, filtering historical data eliminates low-quality data, resulting in a higher quality pool of target data. Furthermore, filtering historical data effectively reduces the amount of data used for iterative training, ensuring and improving the effectiveness of the initial recommendation algorithm's training. Iteratively updating the initial recommendation algorithm with high-quality target data significantly enhances its predictive performance, thereby increasing the accuracy of the search results output by the updated algorithm model.

[0059] In the information processing method provided in Embodiment 1 of this application, in order to ensure the accuracy of the search results output by the target recommendation algorithm, this application defines how to obtain the target recommendation algorithm before processing the search information through the target recommendation algorithm to obtain the search results. Specifically, the method further includes: acquiring multiple historical data information; filtering the multiple historical data information to obtain multiple target data information, and constructing a training sample set based on the multiple target data information. The training sample set includes multiple training samples, which consist of historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information. The first positive sample is the selected search result in the multiple search results set determined based on behavioral feedback data, and the first negative sample is the unselected search result in the multiple search results set determined based on behavioral feedback data. The initial recommendation algorithm is iteratively trained based on the training sample set to obtain the target recommendation algorithm.

[0060] Optionally, the aforementioned historical data can be collected using event tracking technology. This historical data includes, but is not limited to, multiple search result sets output by multiple historical search information and behavioral feedback data on the search results within these sets. After collecting the historical data, to improve the quality of the subsequent training sample set, the historical data is filtered to obtain the aforementioned target data. Finally, the training sample set is constructed using the target data and the historical search information.

[0061] It's important to note that filtering through multiple historical data sets is done to eliminate low-quality data and obtain high-quality target data. Low-quality data primarily refers to data that the recommendation algorithm can correctly predict and has very low uncertainty; for example, the first search result in the algorithm's output has already been selected by the user. High-quality data, on the other hand, represents data where the recommendation algorithm's predictions have greater uncertainty; for example, none of the algorithm's search results have been selected by the user, meaning the user believes that the algorithm's output does not contain the result they want.

[0062] It should be noted that the training sample set consists of multiple training samples, each of which is a triplet: (anchor, positive, negative). The anchor represents the historical search information mentioned above, the positive represents the first positive sample, and the negative represents the first negative sample. The first positive sample is the selected search result, and the first negative sample is the unselected search result. For example, if the initial recommendation algorithm outputs 20 search results based on the user's favorite music, and the third search result is selected by the user, then that third search result is the first positive sample corresponding to the user's favorite music historical search information. All other search results are the first negative samples corresponding to the user's favorite music historical search information.

[0063] In conclusion, the training sample set constructed by filtering the target data information can effectively improve the accuracy of the search results output by the initial recommendation algorithm.

[0064] To more effectively filter multiple historical search data, the information processing method provided in Embodiment 1 of this application filters multiple historical data to obtain multiple target data, including: for each historical data, determining the target matching degree corresponding to the historical data based on behavioral feedback data and / or search results in multiple search result sets, wherein the target matching degree characterizes the degree of matching between the search result set and the historical search information; and filtering the multiple historical data based on the multiple target matching degrees to obtain multiple target data.

[0065] Optionally, the target matching degree corresponding to the historical data information in multiple historical data information can be obtained based on behavioral feedback data and / or search results in multiple search result sets. It should be noted that the target matching degree represents the degree of matching between the search result set and the historical search information, which is simply the accuracy of the search results mentioned above.

[0066] After determining the target matching degree corresponding to the historical data information, the target matching degree is used to filter multiple historical data information to obtain the aforementioned multiple target data information.

[0067] It's important to note that when iteratively training the initial recommendation algorithm, it's preferable to train it using samples with relatively low accuracy to effectively improve its predictive performance. Therefore, when filtering through multiple historical data sets, it's necessary to select data with low target matching as the target data, or in other words, data with high uncertainty, to maximize the improvement of the initial recommendation algorithm.

[0068] In summary, by using target matching degree, target data information can be selected more rationally, thereby improving the predictive performance of the initial recommendation algorithm.

[0069] To enhance the importance and rationality of calculating the target matching degree, in the information processing method provided in Embodiment 1 of this application, for historical data information among multiple historical data information, the target matching degree corresponding to the historical data information is determined based on behavioral feedback data and / or search results in multiple search result sets. The method employed is at least one of the following: For historical data information among multiple historical data information, based on behavioral feedback data, it is determined whether the search results in the search result set corresponding to the historical data information are selected; if the search results in the search result set corresponding to the historical data information are selected, the order of the selected search results in the search result set corresponding to the historical data information is determined, and the target matching degree corresponding to the historical data information is determined based on whether the search results are selected and the order of the selected search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, the target matching degree corresponding to the historical data information is determined based on behavioral feedback data and target search results in multiple search result sets; For historical data information among multiple historical data information, the target matching degree corresponding to the historical data information is obtained based on the category information corresponding to the search results in multiple search result sets.

[0070] Optionally, the information processing method provided in Embodiment 1 of this application proposes the following three methods for calculating the target matching degree:

[0071] Method 1: Screening strategy based on negative feedback.

[0072] Based on the aforementioned behavioral feedback data, it is determined whether the search results in the search results set corresponding to the historical data information were selected, thus obtaining a judgment result. It should be noted that the judgment result indicates whether the search result was selected or not. If the search results in the search results set corresponding to the historical data information were selected, it is also necessary to determine the order of the selected search results within that set. Then, the target matching degree is calculated using whether the search results in the search results set corresponding to the historical data information were selected and the order of the selected search results across multiple search results sets.

[0073] It should be noted that the historical data information of negative feedback can also be determined based on whether the search results are selected and the order in which the selected search results are arranged in multiple search result sets. Then, the historical data information of negative feedback can be directly filtered out as the target data information mentioned above.

[0074] It's important to note that negative feedback refers to data information where the recommendation algorithm makes an incorrect prediction. Taking a recall model as an example, we can assume that if the top-1 prediction result is inconsistent with the user's feedback, it's negative feedback; otherwise, it's positive feedback.

[0075] In an alternative embodiment, negative feedback can also be defined as the top-3 results, where the first three results are all incorrect. The specific definition can be determined based on the task scenario. This strategy assumes that negative feedback samples have greater uncertainty, and if used to construct training samples, they can lead to greater model improvement.

[0076] Method 2: Filtering strategy based on behavioral feedback data and target search results.

[0077] The target matching degree corresponding to the historical data information is determined based on the aforementioned behavioral feedback data and target search results. It should be noted that the target search results are the unselected search results within the search results set corresponding to the historical data information. By using behavioral feedback data and target search results, the uncertainty of the historical data information (i.e., the aforementioned target matching degree) can be accurately assessed.

[0078] Method 3: Filtering strategy based on category information.

[0079] The algorithm retrieves the category information corresponding to the search results in the search results set corresponding to historical data, and then uses the proportion of this category information in the total category set to determine the target matching degree corresponding to the historical data information. It should be noted that the category information here can be tag information; for example, if the recommendation algorithm recommends favorite music, then the category information of the corresponding output search results might be music genre.

[0080] In summary, the target matching degree corresponding to historical data information can be accurately calculated through the three methods described above, thereby achieving the effect of reasonably filtering multiple historical data information.

[0081] It should be noted that any of the methods mentioned above can be used to determine the target matching degree corresponding to historical data information.

[0082] In an optional embodiment, two or more of the above methods can be fused to further improve the quality of the samples and reduce the number of training samples. For example, a selection of samples can be first filtered using a filtering strategy based on behavioral feedback data and target search results, and then further filtered using a filtering strategy based on category information in the selected samples. Numerical fusion filtering can also be used, for example, the following formula (1) can be defined to calculate the target matching degree under the fusion of multiple strategies:

[0083] Uncertainty=α×TRR+β×LE(1)

[0084] Where α and β represent preset parameters. TRR is the target matching degree obtained by using the above-mentioned filtering strategy of behavioral feedback data and target search results, LE is the target matching degree obtained by using the above-mentioned filtering strategy of category information, and the larger the Uncertainty, the greater the final uncertainty.

[0085] In the information processing method provided in Embodiment 1 of this application, for historical data information among multiple historical data information, determining the target matching degree corresponding to the historical data information based on behavioral feedback data and target search results in multiple search result sets includes: determining the target search results in the search result set corresponding to the historical data information based on behavioral feedback data; determining the order of the target search results in the search result set corresponding to the historical data information; and calculating the matching degree between the search result set corresponding to the historical data information and the historical search information based on the order of the order to obtain the target matching degree of the historical data information.

[0086] Optionally, the filtering strategy based on behavioral feedback data and target search results includes: determining the target search results corresponding to the search result set corresponding to historical data information based on the behavioral feedback data. It should be noted that the target search results are the unselected search results in the search result set corresponding to the historical data information. The order of the target search results within the search result set corresponding to the historical data information is then determined. Generally, the higher the order of the unselected search results, the higher the uncertainty of the search result set corresponding to the historical data information, meaning a lower target match. The matching degree between the search result set corresponding to the historical data information and the historical search information is calculated based on the order, ultimately obtaining the target match degree of the historical data information.

[0087] In an optional embodiment, the filtering strategy based on behavioral feedback data and target search results can also be called a TRR (total reciprocal rank) filtering strategy. For the search result set corresponding to historical data information, all incorrect prediction results (i.e., the target search results mentioned above) are first filtered from the top-K search results, using the set {S} i}, 1<=i<=K, where i is the sequence number, indicating that the erroneous sample is ranked i-th in the top-K. Simultaneously, the sequence set {i} of these samples is obtained, and then the sum of the inverse sequences is calculated, as shown in the following formula (2):

[0088]

[0089] The TRR value mentioned above is then used as the target matching degree of the search results set corresponding to the historical data information. The lower the matching degree, the greater the uncertainty of the search results set corresponding to the historical data information.

[0090] In summary, calculating the matching degree between the search results set corresponding to historical data and historical search information by arranging the order improves the rationality of the matching degree calculation.

[0091] In the information processing method provided in Embodiment 1 of this application, for historical data information among multiple historical data information, obtaining the target matching degree corresponding to the historical data information based on the category information corresponding to the search results in multiple search result sets includes: for historical data information among multiple historical data information, obtaining the category information corresponding to the search results in the search result set corresponding to the historical data information; calculating based on the category information to obtain the category entropy corresponding to the search result set in the search result set corresponding to the historical data information; and determining multiple target matching degrees based on the category entropy.

[0092] Optionally, the filtering strategy based on category information includes: obtaining the category information corresponding to the search results in the search result set corresponding to historical data information. It should be noted that the data output by the recommendation algorithm all have corresponding category information, which can also be called the data label. The category entropy (LE) can be calculated based on the category, and then the target matching degree corresponding to the historical data information can be determined through the category entropy.

[0093] In an optional embodiment, assume the category (or label) of the top-K search results is {ai}, 1 <= i <= K, where ai ∈ L, and L represents the total category set. Calculate the proportion {Pn} of all categories appearing in the top-K results in the total category set, where n represents the total number of categories appearing in the top-K results. Then, the formula for calculating the category entropy of the top-K prediction results is shown in formula (3):

[0094] LE = -∑log(P) n (3)

[0095] The category entropy of multiple search result sets is calculated using the above formula (3), and then the category entropy is used as the target matching degree corresponding to the historical data information.

[0096] To further improve the quality of the training sample set, the information processing method provided in Embodiment 1 of this application includes constructing a training sample set based on multiple target data information as follows: filtering multiple target data information based on historical search information to obtain filtered target data information; and constructing a training sample set based on the filtered target data information.

[0097] Optionally, after filtering multiple historical data information in the above manner to obtain the target data information, a training sample set can be directly constructed based on the multiple target data information.

[0098] By randomly combining the aforementioned target data into triplets (three by three), a large number of training samples can be obtained. However, many of these training samples are actually low-quality. At this stage, low-quality samples are defined as those in which the model can easily determine that the first two sentences in a triplet have the same label, while the third label is different. If these low-quality samples constitute the majority, it will not only lead to inefficient training but also result in minimal improvement in model performance.

[0099] In an optional embodiment, multiple target data information are filtered to select more challenging positive and negative samples. It should be noted that challenging positive samples are those with low semantic similarity to the benchmark sample (i.e., the aforementioned historical search information) but consistent labels; challenging negative samples are those with high semantic similarity to the benchmark sample but inconsistent labels. Finally, a training sample set is constructed using the filtered target data information.

[0100] By performing secondary screening on the target data, the quality of the samples in the training sample set can be further improved. The prediction effect of the recommendation algorithm can be effectively improved by using hard positive samples and hard negative samples.

[0101] To improve the accuracy of screening difficult positive and difficult negative samples, the information processing method provided in Embodiment 1 of this application filters multiple target data information based on historical search information to obtain filtered target data information, including: vectorizing the historical search information to obtain a first vector; vectorizing the target data information in the multiple target data information to obtain multiple second vectors; calculating the cosine similarity between the first vector and the multiple second vectors to obtain multiple cosine similarities; and filtering the multiple target data information based on the multiple cosine similarities to obtain filtered target data information.

[0102] Optionally, the historical search information and target data information mentioned above are vectorized to obtain the first vector and the second vector. Then, the cosine similarity between the first vector and multiple second vectors is calculated. The cosine similarity is used to filter multiple target data information to obtain the filtered target data information.

[0103] In an optional embodiment, for positive samples: target data information with a cosine similarity lower than a first set threshold can be directly used as the filtered target data information; for negative samples: target data information with a cosine similarity higher than a second set threshold can be directly used as the filtered target data information.

[0104] For recommendation algorithms, selecting only very difficult negative and positive samples may make the learning process too difficult for the model. Therefore, in an optional embodiment, different similarity thresholds can be set to divide the difficulty into multiple levels, such as very difficult, relatively difficult, and normal difficulty. A portion of positive and negative samples are selected proportionally for each level.

[0105] In conclusion, cosine similarity can be used to reasonably filter target data information, which can effectively improve the sample quality of the training sample set.

[0106] In the information processing method provided in Embodiment 1 of this application, the method of iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm includes: constructing a sample pool based on historical data information other than multiple target data information; determining multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool; and iteratively training the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0107] Optionally, both of the above-mentioned screening stages are static sample selection schemes, completed all at once before training. However, as the recommendation algorithm is continuously trained, the parameters of the algorithm change with each round. Previously difficult positive and negative samples may become relatively easy, while previously average samples may become quite difficult. To further improve the training efficiency of the model, a sample pool is constructed using data from multiple historical data sources excluding multiple target data sources. High-quality negative and positive samples are selected from this pool to construct the positive and negative samples for the training samples in the training sample set. Finally, the initial recommendation algorithm is iteratively trained using multiple second positive samples, multiple second negative samples, and the training samples from the training sample set to obtain the target recommendation algorithm.

[0108] By filtering negative and positive samples from the sample pool, the training effect of the initial recommendation algorithm can be improved.

[0109] In the information processing method provided in Embodiment 1 of this application, determining multiple second positive samples and multiple second negative samples corresponding to training samples in the training sample set from sample data in the sample pool includes: calculating the gradient of sample data in the sample pool according to the initial recommendation algorithm to obtain the gradient vector corresponding to the sample data; clustering according to the gradient vector corresponding to the sample data to obtain multiple initial sample data; and filtering the multiple initial sample data to obtain multiple second positive samples and multiple second negative samples.

[0110] Optionally, taking the acquisition of the second negative sample as an example, this section explains how to select corresponding negative samples from the sample pool. The initial recommendation algorithm is used to calculate the gradient vector of the sample data in the sample pool. It should be noted that the initial recommendation algorithm can calculate the corresponding gradient vector based on the category information of the sample data. After obtaining the gradient vector, the gradient vectors corresponding to the sample data are clustered using the k-means clustering method, selecting m initial sample data (where m is the number of cluster centers in k-means). These m initial sample data are usually more representative than random selection. Then, n initial sample data with the greatest gradient uncertainty are selected from the m initial sample data as the negative sample queue (i.e., the multiple second negative samples mentioned above). It should be noted that gradient uncertainty is represented by the magnitude of the gradient vector; the larger the magnitude, the greater the uncertainty. The second positive sample can also be implemented in the same way. It should be noted that the main difference between the first and second negative samples is that one is the search result corresponding to its own search information, while the other is the search result corresponding to other search information.

[0111] In an optional embodiment, a schematic diagram of selecting the negative sample queue is shown below. Figure 3 As shown, first, a sample pool is constructed, and then m samples are selected from it based on the k-means clustering method (or a variant of k-means). Then, n samples with the greatest gradient uncertainty are selected from the m samples as a negative sample queue. Finally, a training sample triplet includes: anchor, optimal positive, hard negative, and negative sample queue.

[0112] In an alternative embodiment, the following can be employed: Figure 4The flowchart shown automates the iterative training of the initial recommendation algorithm, achieving full automation across six processes: feedback collection, sample selection, sample construction, model retraining, model evaluation, and model relaunch. Based on online model predictions and user behavior feedback, a large number of feedback samples are collected. The next step is to perform high-quality sample selection and efficient training using a three-stage active learning technique. This three-stage active learning technique includes high-quality sample selection, high-quality training sample construction, and high-quality negative sample selection from the sample pool.

[0113] In an alternative embodiment, the following can be employed: Figure 5 The flowchart shown illustrates a three-stage active learning technique. First, it acquires the prediction results of the recommendation algorithm and user behavior feedback data. In the sample selection stage, three strategies are employed for selecting high-quality samples: a negative feedback-based selection strategy, a TRR-based selection strategy, and a label-entropy (LE)-based selection strategy. Then, in the sample construction stage, to further improve sample quality, methods for selecting difficult positive and difficult negative samples are proposed. Finally, in the model training stage, a negative sample queue is obtained through k-means and gradient selection, further improving the iterative training effect of subsequent recommendation algorithms.

[0114] In the information processing method provided in Embodiment 1 of this application, search information is received; the search information is processed by a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information, which includes at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets. The behavioral feedback data is used to characterize whether a search result is selected. The search results are then displayed. This method solves the technical problem that directly updating the algorithm model based on user feedback data after deployment leads to low accuracy of the search results output by the iteratively updated algorithm model. In this solution, filtering historical data information removes low-quality data, resulting in high-quality target data information. Furthermore, filtering historical data information effectively reduces the amount of data used for iterative training, ensuring the effectiveness of iterative training of the initial recommendation algorithm while improving the overall training effect. By iteratively updating the initial recommendation algorithm with high-quality target data, the predictive performance of the initial recommendation algorithm can be effectively improved, thereby increasing the accuracy of the search results output by the updated algorithm model.

[0115] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0117] Example 2

[0118] According to embodiments of this application, an information processing method is also provided, such as... Figure 6 As shown, the information processing method includes:

[0119] Step S601: Receive search information for the target object uploaded by the client;

[0120] Step S602: In the cloud server, the search information is processed by a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm with a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information. The multiple historical data information is filtered according to the target matching degree of the historical data information. The target matching degree represents the degree of matching between the search result set corresponding to the historical data information and the historical search information. The historical data information includes at least: the search result set output by the initial recommendation algorithm for the historical search information and the behavioral feedback data for the search results in the search result set. The behavioral feedback data is used to represent whether the search result is selected.

[0121] Step S603: Send the search results back to the client.

[0122] In the information processing method provided in Embodiment 2 of this application, search information of a target object uploaded by a client is received; the search information is processed in a cloud server using a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information. The filtering of historical data information is based on the target matching degree of the historical data information. The target matching degree characterizes the degree of matching between the search result set corresponding to the historical data information and the historical search information. The historical data information includes at least: the search result set output by the initial recommendation algorithm for the historical search information and behavioral feedback data on the search results in the search result set. The behavioral feedback data is used to characterize whether the search result is selected. Feeding the search results back to the client solves the technical problem that directly updating the algorithm model based on user feedback data after deployment leads to low accuracy of the search results output by the iteratively updated algorithm model. In this approach, by filtering historical data, low-quality data can be eliminated, resulting in multiple high-quality target data sets. Furthermore, filtering historical data effectively reduces the amount of data used for iterative training, ensuring and improving the effectiveness of the initial recommendation algorithm's iterative training. Iteratively updating the initial recommendation algorithm using high-quality target data effectively enhances its predictive performance, thereby increasing the accuracy of the search results output by the updated algorithm model.

[0123] The specific methods for information processing in the cloud server are the same as those in Example 1, and will not be repeated here.

[0124] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0125] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0126] Example 3

[0127] According to embodiments of this application, a processing apparatus for implementing the above-described information processing method is also provided, such as... Figure 7 As shown, the device includes: a receiving unit 701, a processing unit 702, and a display unit 703.

[0128] The receiving unit 701 is used to receive search information;

[0129] Processing unit 702 is used to process search information through a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm with a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information. The multiple historical data information is filtered according to the target matching degree of the historical data information. The target matching degree represents the degree of matching between the search result set corresponding to the historical data information and the historical search information. The historical data information includes at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets. The behavioral feedback data is used to represent whether the search result is selected.

[0130] Display unit 703 is used to display search results.

[0131] In the information processing apparatus provided in Embodiment 3 of this application, a receiving unit 701 receives search information; a processing unit 702 processes the search information using a target recommendation algorithm to obtain search results. The target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information. The multiple target data information is obtained by filtering multiple historical data information. The filtering of multiple historical data information is based on the target matching degree of the historical data information. The target matching degree characterizes the degree of matching between the search result set corresponding to the historical data information and the historical search information. The historical data information includes at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets. The behavioral feedback data is used to characterize whether a search result is selected. A display unit 703 displays the search results. This solves the technical problem that after deploying the algorithm model, directly iterating and updating the algorithm model using user feedback data leads to a relatively low accuracy rate of the search results output by the iteratively updated algorithm model. In this approach, by filtering historical data, low-quality data can be eliminated, resulting in multiple high-quality target data sets. Furthermore, filtering historical data effectively reduces the amount of data used for iterative training, ensuring and improving the effectiveness of the initial recommendation algorithm's iterative training. Iteratively updating the initial recommendation algorithm using high-quality target data effectively enhances its predictive performance, thereby increasing the accuracy of the search results output by the updated algorithm model.

[0132] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the apparatus further includes: an acquisition unit, configured to acquire multiple historical data information before processing the search information through the target recommendation algorithm to obtain search results; a filtering unit, configured to filter the multiple historical data information to obtain multiple target data information, and construct a training sample set based on the multiple target data information, wherein the training sample set includes multiple training samples, the training samples consisting of historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample being the selected search result in the multiple search results set determined based on behavioral feedback data, and the first negative sample being the unselected search result in the multiple search results set determined based on behavioral feedback data; and a training unit, configured to iteratively train the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

[0133] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the filtering unit includes: a first determining subunit, configured to determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the multiple search result sets, wherein the target matching degree characterizes the degree of matching between the multiple search result sets and the multiple historical search information; and a first filtering subunit, configured to filter the multiple historical data information based on the multiple target matching degrees to obtain the multiple target data information.

[0134] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the first determining subunit includes at least one of the following: a first determining module, configured to, for historical data information from multiple historical data information, determine whether a search result in the search result set corresponding to the historical data information is selected based on the behavioral feedback data, and if a search result in the search result set corresponding to the historical data information is selected, determine the order of the selected search result in the search result set corresponding to the historical data information, and determine the target matching degree corresponding to the historical data information based on whether the search result is selected and the order of the selected search result in the search result set corresponding to the historical data information; a second determining module, configured to, for historical data information from multiple historical data information, determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search result in the search result set corresponding to the historical data information; and a third determining module, configured to, for historical data information from multiple historical data information, obtain the target matching degree corresponding to the historical data information based on the category information corresponding to the search result in the search result set corresponding to the historical data information.

[0135] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the second determining module includes: a first determining submodule, used to determine, based on the behavioral feedback data, the target search result in the search results corresponding to the historical data information for a plurality of historical data information; a second determining submodule, used to determine the order of the target search result in the search results set corresponding to the historical data information; and a first calculating submodule, used to calculate the degree of matching between the search results set corresponding to the historical data information and the historical search information based on the order of matching, to obtain the target matching degree of the historical data information.

[0136] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the third determining module includes: an acquisition submodule, used to acquire category information corresponding to the search results in the search results set corresponding to the historical data information; a second calculation submodule, used to perform calculations based on the category information to obtain the category entropy corresponding to the search results set corresponding to the historical data information; and a third determining submodule, used to determine the target matching degree corresponding to the historical data information based on the category entropy.

[0137] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the filtering unit further includes: a second filtering subunit, used to filter the plurality of target data information based on the historical search information to obtain filtered target data information; and a construction subunit, used to construct the training sample set based on the filtered target data information.

[0138] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the second filtering subunit includes: a first processing module, used to vectorize the historical search information to obtain a first vector; a second processing module, used to vectorize the target data information in the plurality of target data information to obtain a plurality of second vectors; a first calculation module, used to calculate the cosine similarity between the first vector and the plurality of second vectors to obtain a plurality of cosine similarities; and a first filtering module, used to filter the plurality of target data information based on the plurality of cosine similarities to obtain filtered target data information.

[0139] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the training unit includes: a construction subunit, configured to construct a sample pool based on historical data information other than the multiple target data information; a second determination subunit, configured to determine multiple second positive samples and multiple second negative samples corresponding to training samples in the training sample set from the sample data in the sample pool; and a training subunit, configured to iteratively train the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0140] Optionally, in the information processing apparatus provided in Embodiment 3 of this application, the second determining subunit includes: a second calculation module, used to calculate the gradient of the sample data in the sample pool according to the initial recommendation algorithm, to obtain the gradient vector corresponding to the sample data; a clustering module, used to cluster according to the gradient vector corresponding to the sample data, to obtain multiple initial sample data; and a second filtering module, used to filter the multiple initial sample data to obtain the multiple second positive samples and the multiple second negative samples.

[0141] It should be noted that the receiving unit 701, processing unit 702, and display unit 703 mentioned above correspond to steps S201 to S203 in Embodiment 1. The two modules and their corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the device, can run in the computer terminal 10 provided in Embodiment 1.

[0142] It should be noted that the preferred implementation schemes involved in the above embodiments of this application are the same as the schemes, application scenarios and implementation processes provided in Embodiment 1, but are not limited to the schemes provided in Embodiment 1.

[0143] Example 4

[0144] Embodiments of this application may provide a computer terminal, which may be any computer terminal device in a group of computer terminals. Optionally, in this embodiment, the aforementioned computer terminal may also be replaced by a mobile terminal or other terminal device.

[0145] Optionally, in this embodiment, the computer terminal may be located in at least one of a plurality of network devices in a computer network.

[0146] In this embodiment, the computer terminal described above can execute program code for the following steps in the information processing method: receiving search information; processing the search information using a target recommendation algorithm to obtain search results, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information, the multiple target data information being obtained by filtering multiple historical data information, and filtering the multiple historical data information based on the target matching degree of the historical data information, the target matching degree representing the degree of matching between the search result set corresponding to the historical data information and the historical search information, the historical data information including at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets, wherein the behavioral feedback data is used to represent whether the search result is selected; and displaying the search results.

[0147] The aforementioned computer terminal can execute program code for the following steps in the information processing method: Before processing the search information using a target recommendation algorithm to obtain search results, the method further includes: acquiring the plurality of historical data information; filtering the plurality of historical data information to obtain the plurality of target data information, and constructing the training sample set based on the plurality of target data information, wherein the training sample set includes a plurality of training samples, the training samples consisting of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample being the selected search result in the plurality of search results set determined based on the behavioral feedback data, and the first negative sample being the unselected search result in the plurality of search results set determined based on the behavioral feedback data; iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

[0148] The aforementioned computer terminal can execute program code for the following steps in the information processing method: filtering the plurality of historical data information to obtain a plurality of target data information includes: for historical data information among the plurality of historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the search results set corresponding to the historical data information, wherein the target matching degree characterizes the degree of matching between the search results set and the historical search information; filtering the plurality of historical data information based on the target matching degree to obtain the plurality of target data information.

[0149] The aforementioned computer terminal can execute program code for the following steps in the information processing method: For historical data information among multiple historical data information, based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, determine the target matching degree corresponding to the historical data information, using at least one of the following methods: For historical data information among multiple historical data information, based on the behavioral feedback data, determine whether the search results in the search result set corresponding to the historical data information are selected; and if the search results in the search result set corresponding to the historical data information are selected, determine the order of the selected search results in the search result set corresponding to the historical data information, and determine the target matching degree corresponding to the historical data information based on whether the search results are selected and the order of the selected search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, obtain the target matching degree corresponding to the historical data information based on the category information corresponding to the target search results in the search result set corresponding to the historical data information.

[0150] The aforementioned computer terminal can execute program code for the following steps in the information processing method: For historical data information among multiple historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search result in the search result set corresponding to the historical data information includes: For historical data information among multiple historical data information, determining the target search result in the search result set corresponding to the historical data information based on the behavioral feedback data; determining the order of the target search result in the search result set corresponding to the historical data information; calculating the matching degree between the search result set corresponding to the historical data information and the historical search information based on the order of the order, to obtain the target matching degree of the historical data information.

[0151] The aforementioned computer terminal can execute program code for the following steps in the information processing method: For historical data information among multiple historical data information, obtaining the target matching degree corresponding to the historical data information based on the category information corresponding to the target search result in the search result set corresponding to the historical data information includes: obtaining the category information corresponding to the search result in the search result set corresponding to the historical data information; calculating based on the category information to obtain the category entropy corresponding to the search result set in the search result set corresponding to the historical data information; and determining the target matching degree corresponding to the historical data information based on the category entropy.

[0152] The aforementioned computer terminal can execute program code for the following steps in the information processing method: constructing a training sample set based on the plurality of target data information includes: filtering the plurality of target data information based on the historical search information to obtain filtered target data information; and constructing the training sample set based on the filtered target data information.

[0153] The aforementioned computer terminal can execute program code for the following steps in the information processing method: filtering the multiple target data information based on the historical search information to obtain filtered target data information includes: vectorizing the historical search information to obtain a first vector; vectorizing the target data information in the multiple target data information to obtain multiple second vectors; calculating the cosine similarity between the first vector and the multiple second vectors to obtain multiple cosine similarities; and filtering the multiple target data information based on the multiple cosine similarities to obtain filtered target data information.

[0154] The aforementioned computer terminal can execute program code for the following steps in the information processing method: iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm, including: constructing a sample pool based on historical data information other than the multiple target data information; determining multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool; iteratively training the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0155] The aforementioned computer terminal can execute program code for the following steps in the information processing method: determining multiple second positive samples and multiple second negative samples corresponding to training samples in the training sample set from sample data in the sample pool includes: calculating the gradient of sample data in the sample pool according to the initial recommendation algorithm to obtain the gradient vector corresponding to the sample data; clustering according to the gradient vector corresponding to the sample data to obtain multiple initial sample data; and filtering the multiple initial sample data to obtain the multiple second positive samples and the multiple second negative samples.

[0156] Optionally, Figure 8 This is a structural block diagram of a computer terminal according to an embodiment of this application. Figure 8 As shown, the computer terminal 10 may include: one or more ( Figure 8(Only one is shown in the image) Processor 102 and memory 104. The computer terminal 10 may also include a storage controller to control and manage the memory 104; the computer terminal 10 may also include a peripheral interface to connect to a radio frequency module, an audio module, and a display screen, etc.

[0157] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the information processing method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned information processing method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0158] The processor can invoke information and applications stored in memory via a transmission device to perform the following steps: receiving search information; processing the search information using a target recommendation algorithm to obtain search results, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information, the multiple target data information being obtained by filtering multiple historical data information, and the multiple historical data information being filtered based on the target matching degree of the historical data information, the target matching degree representing the degree of matching between the search result set corresponding to the historical data information and the historical search information, the historical data information including at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets, wherein the behavioral feedback data is used to represent whether the search result is selected; and displaying the search results.

[0159] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: before processing the search information using a target recommendation algorithm to obtain search results, the method further includes: acquiring the plurality of historical data information; filtering the plurality of historical data information to obtain the plurality of target data information, and constructing the training sample set based on the plurality of target data information, wherein the training sample set includes a plurality of training samples, the training samples consisting of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample being the selected search result in the plurality of search results set determined based on the behavioral feedback data, and the first negative sample being the unselected search result in the plurality of search results set determined based on the behavioral feedback data; iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

[0160] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: filtering the plurality of historical data information to obtain a plurality of target data information includes: for historical data information among the plurality of historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, wherein the target matching degree characterizes the degree of matching between the search result set and the historical search information; filtering the plurality of historical data information based on the target matching degree to obtain the plurality of target data information.

[0161] The processor can invoke information and applications stored in memory via a transmission device to perform the following steps: For historical data information among multiple historical data information, determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, using at least one of the following methods: For historical data information among multiple historical data information, determine whether the search results in the search result set corresponding to the historical data information are selected based on the behavioral feedback data, and if the search results in the search result set corresponding to the historical data information are selected, determine the order of the selected search results in the search result set corresponding to the historical data information, and determine the target matching degree corresponding to the historical data information based on whether the search results are selected and the order of the selected search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, obtain the target matching degree corresponding to the historical data information based on the category information corresponding to the target search results in the search result set corresponding to the historical data information.

[0162] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: For historical data information among multiple historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search results in the search results set corresponding to the historical data information includes: For historical data information among multiple historical data information, determining the target search results in the search results set corresponding to the historical data information based on the behavioral feedback data; determining the order of the target search results in the search results set corresponding to the historical data information; calculating the matching degree between the search results set corresponding to the historical data information and the historical search information based on the order of the order, to obtain the target matching degree of the historical data information.

[0163] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: For historical data information among multiple historical data information, obtaining the target matching degree corresponding to the historical data information based on the category information corresponding to the target search result in the search result set corresponding to the historical data information includes: obtaining the category information corresponding to the search result in the search result set corresponding to the historical data information; calculating the category entropy corresponding to the search result set in the search result set corresponding to the historical data information based on the category information; and determining the target matching degree corresponding to the historical data information based on the category entropy.

[0164] The processor can invoke information and applications stored in the memory through the transmission device to perform the following steps: constructing a training sample set based on the plurality of target data information includes: filtering the plurality of target data information based on the historical search information to obtain filtered target data information; and constructing the training sample set based on the filtered target data information.

[0165] The processor can invoke information and application programs stored in the memory via a transmission device to perform the following steps: filtering the plurality of target data information based on the historical search information to obtain filtered target data information includes: vectorizing the historical search information to obtain a first vector; vectorizing the target data information in the plurality of target data information to obtain a plurality of second vectors; calculating the cosine similarity between the first vector and the plurality of second vectors to obtain a plurality of cosine similarities; and filtering the plurality of target data information based on the plurality of cosine similarities to obtain filtered target data information.

[0166] The processor can invoke information and application programs stored in the memory via a transmission device to perform the following steps: iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm, including: constructing a sample pool based on historical data information other than the multiple target data information; determining multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool; iteratively training the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0167] The processor can invoke information and application programs stored in the memory via a transmission device to perform the following steps: determining multiple second positive samples and multiple second negative samples corresponding to training samples in the training sample set from the sample data in the sample pool includes: calculating the gradient of the sample data in the sample pool according to the initial recommendation algorithm to obtain the gradient vector corresponding to the sample data; clustering according to the gradient vector corresponding to the sample data to obtain multiple initial sample data; and filtering the multiple initial sample data to obtain the multiple second positive samples and the multiple second negative samples.

[0168] Those skilled in the art will understand that Figure 8 The structure shown is for illustrative purposes only. The computer terminal can also be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, and other terminal devices. Figure 8This does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include components that are more advanced than those described above. Figure 8 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 8 The different configurations shown.

[0169] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0170] Example 5

[0171] Embodiments of this application also provide a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the information processing method provided in Embodiment 1.

[0172] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0173] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: receiving search information; processing the search information using a target recommendation algorithm to obtain search results, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm using a training sample set constructed from multiple target data information, the multiple target data information being obtained by filtering multiple historical data information, the multiple historical data information being filtered according to the target matching degree of the historical data information, the target matching degree representing the degree of matching between the search result set corresponding to the historical data information and the historical search information, the historical data information including at least: multiple search result sets output by the initial recommendation algorithm for multiple historical search information and behavioral feedback data for the search results in the multiple search result sets, wherein the behavioral feedback data is used to represent whether the search result is selected; and displaying the search results.

[0174] The aforementioned storage medium is configured to store program code for performing the following steps: before processing the search information using a target recommendation algorithm to obtain search results, the method further includes: acquiring the plurality of historical data information; filtering the plurality of historical data information to obtain the plurality of target data information, and constructing the training sample set based on the plurality of target data information, wherein the training sample set includes a plurality of training samples, the training samples consisting of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample being the selected search result in the plurality of search results set determined based on the behavioral feedback data, and the first negative sample being the unselected search result in the plurality of search results set determined based on the behavioral feedback data; iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

[0175] The aforementioned storage medium is configured to store program code for performing the following steps: filtering the plurality of historical data information to obtain a plurality of target data information, including: for historical data information among the plurality of historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, wherein the target matching degree characterizes the degree of matching between the search result set and the historical search information; filtering the plurality of historical data information based on the target matching degree to obtain the plurality of target data information.

[0176] The aforementioned storage medium is configured to store program code for performing the following steps: For historical data information among multiple historical data information, based on the behavioral feedback data and / or the search results in the search result set corresponding to the historical data information, determine the target matching degree corresponding to the historical data information, using at least one of the following methods: For historical data information among multiple historical data information, based on the behavioral feedback data, determine whether the search results in the search result set corresponding to the historical data information are selected; and if the search results in the search result set corresponding to the historical data information are selected, determine the order of the selected search results in the search result set corresponding to the historical data information, and determine the target matching degree corresponding to the historical data information based on whether the search results are selected and the order of the selected search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, determine the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search results in the search result set corresponding to the historical data information; For historical data information among multiple historical data information, obtain the target matching degree corresponding to the historical data information based on the category information corresponding to the target search results in the search result set corresponding to the historical data information.

[0177] The aforementioned storage medium is configured to store program code for performing the following steps: For historical data information among multiple historical data information, determining the target matching degree corresponding to the historical data information based on the behavioral feedback data and the target search result in the search result set corresponding to the historical data information includes: For historical data information among multiple historical data information, determining the target search result in the search result set corresponding to the historical data information based on the behavioral feedback data; determining the order of the target search result in the search result set corresponding to the historical data information; calculating the matching degree between the search result set corresponding to the historical data information and the historical search information based on the order of the order, to obtain the target matching degree of the historical data information.

[0178] The aforementioned storage medium is configured to store program code for performing the following steps: For historical data information in multiple historical data information, obtaining the target matching degree corresponding to the historical data information based on the category information corresponding to the target search result in the search result set corresponding to the historical data information includes: obtaining the category information corresponding to the search result in the search result set corresponding to the historical data information; calculating the category entropy corresponding to the search result set in the search result set corresponding to the historical data information based on the category information; and determining the target matching degree corresponding to the historical data information based on the category entropy.

[0179] The aforementioned storage medium is configured to store program code for performing the following steps: constructing a training sample set based on the plurality of target data information includes: filtering the plurality of target data information based on the historical search information to obtain filtered target data information; and constructing the training sample set based on the filtered target data information.

[0180] The aforementioned storage medium is configured to store program code for performing the following steps: filtering the plurality of target data information based on the historical search information to obtain filtered target data information includes: vectorizing the historical search information to obtain a first vector; vectorizing the target data information in the plurality of target data information to obtain a plurality of second vectors; calculating the cosine similarity between the first vector and the plurality of second vectors to obtain a plurality of cosine similarities; and filtering the plurality of target data information based on the plurality of cosine similarities to obtain filtered target data information.

[0181] The aforementioned storage medium is configured to store program code for performing the following steps: iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm, including: constructing a sample pool based on historical data information other than the multiple target data information; determining multiple second positive samples and multiple second negative samples corresponding to the training samples in the training sample set from the sample data in the sample pool; iteratively training the initial recommendation algorithm based on the multiple second positive samples, the multiple second negative samples, and the training samples in the training sample set to obtain the target recommendation algorithm.

[0182] The aforementioned storage medium is configured to store program code for performing the following steps: determining multiple second positive samples and multiple second negative samples corresponding to training samples in the training sample set from sample data in the sample pool includes: calculating the gradient of the sample data in the sample pool according to the initial recommendation algorithm to obtain the gradient vector corresponding to the sample data; clustering according to the gradient vector corresponding to the sample data to obtain multiple initial sample data; and filtering the multiple initial sample data to obtain the multiple second positive samples and the multiple second negative samples.

[0183] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0184] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0185] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0186] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0187] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0188] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0189] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An information processing method characterized by comprising: The method comprises: receiving search information; processing the search information through a target recommendation algorithm to obtain a search result, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm based on a training sample set constructed from a plurality of target data information, the plurality of target data information is obtained by screening a plurality of historical data information, the plurality of historical data information is screened based on a target matching degree of the historical data information, the target matching degree represents a matching degree between a search result set corresponding to the historical data information and historical search information, the historical data information at least includes a plurality of search result sets output by the initial recommendation algorithm based on a plurality of historical search information and behavior feedback data for search results in the plurality of search result sets, wherein the behavior feedback data is used to represent whether the search result is selected; displaying the search result; wherein a sample pool is constructed based on historical data information other than the plurality of target data information from the plurality of historical data information; determining a plurality of second positive samples and a plurality of second negative samples corresponding to a training sample in the training sample set from sample data in the sample pool, wherein the training sample set includes a plurality of training samples, the training sample is composed of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample is a selected search result in the plurality of search result sets determined based on the behavior feedback data, and the first negative sample is a search result not selected in the plurality of search result sets determined based on the behavior feedback data; iteratively training the initial recommendation algorithm based on the plurality of second positive samples, the plurality of second negative samples, and the training sample in the training sample set to obtain the target recommendation algorithm.

2. The method of claim 1, wherein, Before processing the search information through the target recommendation algorithm to obtain the search result, the method further comprises: obtaining the plurality of historical data information; screening the plurality of historical data information to obtain the plurality of target data information, and constructing the training sample set based on the plurality of target data information; iteratively training the initial recommendation algorithm based on the training sample set to obtain the target recommendation algorithm.

3. The method of claim 2, wherein, Screening the plurality of historical data information to obtain the plurality of target data information comprises: for historical data information in the plurality of historical data information, determining a target matching degree corresponding to the historical data information based on the behavior feedback data and / or search results in a search result set corresponding to the historical data information, wherein the target matching degree represents a matching degree between the search result set and the historical search information; screening the plurality of historical data information based on the target matching degree to obtain the plurality of target data information.

4. The method of claim 3, wherein, For historical data information in the plurality of historical data information, the target matching degree corresponding to the historical data information is determined based on the behavior feedback data and / or the search results in the plurality of search result sets, and the method adopts at least one of the following ways: For the historical data information in the plurality of historical data information, according to the behavior feedback data, it is determined whether the search result in the search result set corresponding to the historical data information is selected, and if the search result in the search result set corresponding to the historical data information is selected, it is determined the arrangement order of the selected search result in the search result set corresponding to the historical data information, and according to whether the search result is selected and the arrangement order of the selected search result in the plurality of search result sets, the target matching degree corresponding to the historical data information is determined; For the historical data information in the plurality of historical data information, according to the behavior feedback data and the target search result in the search result set corresponding to the historical data information, the target matching degree corresponding to the historical data information is determined; For the historical data information in the plurality of historical data information, according to the category information corresponding to the search result in the search result set corresponding to the historical data information, the target matching degree corresponding to the historical data information is obtained.

5. The method of claim 4, wherein, For the historical data information in the plurality of historical data information, according to the behavior feedback data and the target search result in the search result set corresponding to the historical data information, the target matching degree corresponding to the historical data information is determined, comprising: For the historical data information in the plurality of historical data information, according to the behavior feedback data, the target search result in the search result set corresponding to the historical data information is determined; The arrangement order of the target search result in the search result set corresponding to the historical data information is determined; According to the arrangement order, the matching degree between the search result set corresponding to the historical data information and the historical search information is calculated to obtain the target matching degree of the historical data information.

6. The method of claim 4, wherein For the historical data information in the plurality of historical data information, according to the category information corresponding to the search result in the search result set corresponding to the historical data information, the target matching degree corresponding to the historical data information is obtained, comprising: For the historical data information in the plurality of historical data information, the category information corresponding to the search result in the search result set corresponding to the historical data information is obtained; According to the category information, the category entropy corresponding to the search result set in the search result set corresponding to the historical data information is calculated; According to the category entropy, the target matching degree of the historical data information is determined.

7. The method of claim 2, wherein, According to the plurality of target data information, the training sample set is constructed, comprising: According to the historical search information, the plurality of target data information is screened to obtain screened target data information; According to the screened target data information, the training sample set is constructed.

8. The method of claim 7, wherein, According to the historical search information, the plurality of target data information is screened to obtain screened target data information, comprising: The historical search information is vectorized to obtain a first vector; The target data information in the plurality of target data information is vectorized to obtain a plurality of second vectors; The cosine similarity of the first vector and the plurality of second vectors is calculated to obtain a plurality of cosine similarities; Screening the target data information according to the cosine similarities, to obtain screened target data information.

9. The method of claim 1, wherein, Determining, from sample data in the sample pool, second positive samples and second negative samples corresponding to training samples in the training sample set includes: According to the initial recommendation algorithm, the gradient of the sample data in the sample pool is calculated to obtain the gradient vector corresponding to the sample data; According to the gradient vector corresponding to the sample data, a plurality of initial sample data are obtained by clustering; Screening the plurality of initial sample data to obtain the plurality of second positive samples and the plurality of second negative samples.

10. A method of recommending information, characterized by, Comprise: Receiving search information of a target object uploaded by a client; Processing the search information by a target recommendation algorithm in a cloud server to obtain a search result, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm by a training sample set constructed by a plurality of target data information, the plurality of target data information is obtained by screening a plurality of historical data information, the plurality of historical data information is screened according to a target matching degree of the historical data information, the target matching degree represents a matching degree between a search result set corresponding to the historical data information and historical search information, and the historical data information at least includes: a search result set output by the initial recommendation algorithm for historical search information and behavior feedback data for search results in the search result set, wherein the behavior feedback data is used to represent whether the search result is selected; Feedback the search result to the client; Wherein, according to a plurality of historical data information except a plurality of target data information, a sample pool is constructed; From the sample data in the sample pool, a plurality of second positive samples and a plurality of second negative samples corresponding to training samples in the training sample set are determined, wherein the training sample set includes a plurality of training samples, the training sample is composed of the historical search information, the first positive sample corresponding to the historical search information and the first negative sample corresponding to the historical search information, the first positive sample is a selected search result in the plurality of search results according to the behavior feedback data, and the first negative sample is a search result not selected in the plurality of search results according to the behavior feedback data. According to the plurality of second positive samples, the plurality of second negative samples and the training samples in the training sample set, the initial recommendation algorithm is iteratively trained to obtain the target recommendation algorithm.

11. An information processing apparatus comprising: Comprise: A receiving unit for receiving search information; The processing unit is configured to process the search information by using a target recommendation algorithm to obtain a search result, wherein the target recommendation algorithm is obtained by iteratively training an initial recommendation algorithm based on a training sample set constructed from a plurality of target data information, the plurality of target data information is obtained by screening a plurality of historical data information, and the historical data information at least includes a plurality of search result sets output by the initial recommendation algorithm based on a plurality of historical search information and behavior feedback data corresponding to the search results in the plurality of search result sets, wherein the behavior feedback data is used to represent whether the search result is selected; The display unit is configured to display the search result. The device is further configured to construct a sample pool based on the historical data information other than the plurality of target data information in the plurality of historical data information; The plurality of second positive samples and the plurality of second negative samples corresponding to the training sample in the training sample set are determined from the sample data in the sample pool, wherein the training sample set includes a plurality of training samples, the training sample is composed of the historical search information, a first positive sample corresponding to the historical search information, and a first negative sample corresponding to the historical search information, the first positive sample is a selected search result in the plurality of search result sets determined based on the behavior feedback data, and the first negative sample is a non-selected search result in the plurality of search result sets determined based on the behavior feedback data; The initial recommendation algorithm is iteratively trained based on the plurality of second positive samples, the plurality of second negative samples, and the training sample in the training sample set to obtain the target recommendation algorithm.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a stored program, wherein the program controls the device where the storage medium is located to execute the information processing method in any one of claims 1 to 9 when the program is running.

13. An electronic device, comprising: The computer readable storage medium includes a stored program, wherein the program controls the device where the storage medium is located to execute the information processing method in any one of claims 1 to 9 when the program is running. The memory stores an executable program; The processor is configured to run the program, wherein the program executes the information processing method in any one of claims 1 to 9 when the program is running.

Images