A data recommendation method and system based on time series features

Abstract

The present application belongs to the technical field of network data recommendation, and particularly relates to a data recommendation method and system based on time sequence characteristics. The present application constructs multiple parallel time periods by pre-processing and quantitatively mapping user online behavior data, analyzes active periods and blank periods according to time distribution characteristics, and then performs clustering analysis on the online behavior of the user to obtain multiple user behavior categories. This method not only improves the efficiency and accuracy of data processing, but also recommends content that is more in line with the interests and needs of the user according to the specific behavior characteristics of the user. Meanwhile, a feedback scoring mechanism is introduced, the feedback score of the user on the recommended content is obtained in real time, and the optimal recommended content and the alternative recommended content are dynamically adjusted to adapt to the changes in user behavior, thereby further improving the accuracy and personalization of the recommendation.

Description

Technical Field

[0001] The present invention belongs to the technical field of network data recommendation, and in particular relates to a data recommendation method and system based on time series features. Background Art

[0002] With the rapid development of the Internet, the amount of network data has exploded. Users are faced with a huge amount of data to choose from. How to quickly and accurately find content that meets their needs from this data has become an urgent problem to be solved. Based on this, data recommendation methods have emerged. They can recommend personalized data that meets the needs of users based on their browsing habits, thereby reducing the time users spend on screening data.

[0003] However, existing data recommendation methods are mostly based on users' static features and cannot accurately capture and reflect the dynamic changes in user behavior. Therefore, in the data recommendation process, users' behavior patterns, interest preferences, and needs are often not accurately understood and predicted, resulting in low accuracy and personalization of recommendations. To solve this problem, the present invention proposes a data recommendation method based on time series features, which aims to achieve more accurate and personalized data recommendations by capturing and analyzing the dynamic changes in users' online behavior. Summary of the Invention

[0004] The purpose of this invention is to provide a data recommendation method and system based on time series characteristics, which can accurately understand and predict user behavior patterns, interest preferences and needs by capturing and analyzing the dynamic changes of user online behavior, thereby improving the accuracy and personalization of data recommendations.

[0005] The technical solutions adopted by the present invention are as follows:

[0006] A data recommendation method based on time series features, comprising:

[0007] Preprocessing the acquired online behavior data of the user, and performing quantitative mapping processing on the preprocessed online behavior data to obtain benchmark feature parameters;

[0008] Obtaining active segments from each parallel time period according to the occurrence time of each of the benchmark characteristic parameters;

[0009] Performing time distribution feature analysis based on the active segments to extract time series features;

[0010] Performing cluster analysis on the user's online behavior based on the user's active duration in the time series feature to obtain multiple user behavior categories, and classifying the recommendation data according to the user behavior categories to obtain preferred recommended content and alternative recommended content;

[0011] The preferred recommended content and the alternative recommended content are dynamically adjusted according to the feedback scores of each of the user behavior categories, and data recommendation is performed based on the dynamically adjusted preferred recommended content and the alternative recommended content.

[0012] In a preferred embodiment, the step of preprocessing the acquired online behavior data of the user and performing quantitative mapping processing on the preprocessed online behavior data to obtain the benchmark feature parameters includes:

[0013] Acquire online behavior data of all users, and perform data cleaning on the online behavior data to remove duplicate and abnormal records in the online behavior data to obtain cleaned data;

[0014] Acquiring feature extraction conditions, and performing feature extraction on the cleaned data according to the feature extraction conditions to obtain demand features, wherein the feature extraction conditions include the user's operation records, browsing records, and purchase records at different time points;

[0015] All the required characteristics are converted into quantitative values ​​of the same standard through quantitative mapping, and the quantitative values ​​are recorded as benchmark characteristic parameters.

[0016] In a preferred embodiment, the step of obtaining active segments from each parallel time period according to the occurrence time of each reference characteristic parameter includes:

[0017] Perform backtracking offset based on the current time node and output the monitoring period based on the backtracking offset result;

[0018] The monitoring period is divided into a plurality of equally spaced time periods, and the plurality of equally spaced time periods are recorded as parallel time periods;

[0019] According to the occurrence time nodes of each of the benchmark characteristic parameters, the number of occurrences and duration of the user behavior in each of the parallel time periods are counted, and the duration is recorded as the user online period, and the period in which no user behavior occurs is directly recorded as a blank period;

[0020] The length of the online time period of the user is screened to obtain active segments and sporadic segments, and the sporadic segments are merged with adjacent blank segments to obtain adjusted blank segments and active segments.

[0021] In a preferred solution, the step of filtering the length of the user's online time period to obtain active segments and sporadic segments, and merging the sporadic segments with adjacent blank segments to obtain adjusted blank segments and active segments includes:

[0022] Obtaining all online time periods of the users, and recording the duration of the online time periods of the users as parameters to be screened;

[0023] Obtaining a screening threshold, and comparing the screening threshold with the parameter to be screened;

[0024] If the parameter to be screened is greater than or equal to the screening threshold, marking the user's online period as an active period;

[0025] If the parameter to be screened is less than the screening threshold, the user's online period is marked as an occasional period, and when the occasional period is continuous, it is merged into an active segment; otherwise, it is merged with an adjacent blank segment.

[0026] In a preferred embodiment, the step of performing time distribution feature analysis based on the active segment to extract time series features includes:

[0027] Obtaining the active segments in each of the parallel time periods, and calculating the total duration of the active segments and the total duration of the blank segments in each of the parallel time periods;

[0028] The ratio of the total duration of the active segment to the duration of the parallel period is used as the user's activity rate in the parallel period, and the activity rates of all parallel periods are counted to obtain an activity rate time series;

[0029] Counting the number of active users in each of the parallel time periods to obtain an active frequency time series;

[0030] The activity rate time series and the activity frequency time series are used as the time series features of the user in the time period.

[0031] In a preferred embodiment, the step of performing cluster analysis on the user's online behavior based on the user's active duration in the time series feature to obtain multiple user behavior categories includes:

[0032] Obtaining the active duration of each active segment in each of the parallel time periods, and the proportion of the active duration of each active segment in the corresponding parallel time period, and recording the proportion as a condition parameter;

[0033] Acquire a clustering interval, wherein a plurality of clustering intervals are provided, and each clustering interval corresponds to a user behavior category;

[0034] The condition parameters are compared with the cluster intervals one by one, and the user behavior category under each active segment is output based on the comparison results.

[0035] In a preferred solution, the step of classifying the recommendation data according to the user behavior category to obtain preferred recommendation content and alternative recommendation content includes:

[0036] Obtaining the browsing time and purchase record of each recommended content by the user under each user behavior category, and then marking the browsing time and purchase record as the first evaluation parameter and the second evaluation parameter respectively;

[0037] Obtaining an evaluation function, inputting the first evaluation parameter and the second evaluation parameter into the evaluation function, and calibrating an output result of the evaluation function as an evaluation score;

[0038] Obtaining an evaluation threshold, and comparing the evaluation threshold with the evaluation score;

[0039] If the evaluation score is greater than or equal to the evaluation threshold, the corresponding content to be recommended is marked as preferred recommended content;

[0040] If the evaluation score is less than the evaluation threshold, the corresponding content to be recommended is marked as candidate recommendation content.

[0041] In a preferred embodiment, the step of dynamically adjusting the preferred recommended content and the candidate recommended content based on the feedback scores of each of the user behavior categories, and performing data recommendation based on the dynamically adjusted preferred recommended content and the candidate recommended content, includes:

[0042] Constructing a feedback statistical period, and collecting feedback scores of users on recommended content under each of the user behavior categories within the feedback statistical period;

[0043] After the feedback statistics period ends, a standard function is obtained, the feedback score is input into the standard function, and the output value of the standard function is recorded as the recommendation optimization score;

[0044] The preferred recommended content and the alternative recommended content are re-ranked according to the recommendation optimization score, and data recommendation is performed according to the ranking result of the preferred recommended content and the alternative recommended content.

[0045] The present invention also provides a data recommendation system based on time series features, comprising:

[0046] A data acquisition module, configured to preprocess the acquired online behavior data of the user and perform quantitative mapping on the preprocessed online behavior data to obtain benchmark feature parameters;

[0047] a time period division module, the time period division module being configured to obtain active segments from each parallel time period according to the occurrence time of each of the reference characteristic parameters;

[0048] A feature analysis module, configured to perform time distribution feature analysis based on the active segment and extract time series features;

[0049] A cluster analysis module, configured to perform cluster analysis on the user's online behavior based on the user's active duration in the time series feature to obtain multiple user behavior categories, and to classify the recommendation data according to the user behavior categories to obtain preferred recommended content and alternative recommended content;

[0050] A dynamic adjustment module is used to dynamically adjust the preferred recommended content and the alternative recommended content according to the feedback scores of each of the user behavior categories, and to perform data recommendation based on the dynamically adjusted preferred recommended content and the alternative recommended content.

[0051] And, an electronic device, comprising:

[0052] at least one processor;

[0053] and a memory communicatively coupled to the at least one processor;

[0054] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the above-mentioned data recommendation method based on time series features.

[0055] The technical effects achieved by the present invention are:

[0056] The present invention constructs multiple parallel time periods by preprocessing and quantitatively mapping user online behavior data, and analyzes active segments and blank segments based on time distribution characteristics, and then performs cluster analysis on users' online behaviors to obtain multiple user behavior categories. This method not only improves the efficiency and accuracy of data processing, but also recommends content that better suits users' interests and needs based on their specific behavioral characteristics. At the same time, it also introduces a feedback scoring mechanism. By collecting users' feedback scores on recommended content, and under the premise of sufficient feedback data, the preferred recommended content and alternative recommended content can be periodically and dynamically adjusted to adapt to changes in user behavior, further improving the accuracy and personalization of recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic flow chart of the method of the present invention;

[0058] Figure 2 It is a schematic diagram of the system modules of the present invention;

[0059] Figure 3 It is a schematic structural diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0060] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it constitute a separate or selective embodiment that is mutually exclusive of other embodiments.

[0063] See also Figure 1 As shown, the present invention provides a data recommendation method based on time series features, comprising:

[0064] S1. Preprocess the acquired online behavior data of the user and perform quantitative mapping on the preprocessed online behavior data to obtain benchmark feature parameters.

[0065] S2. Obtain active segments from each parallel time period based on the occurrence time of each benchmark characteristic parameter;

[0066] S3. Analyze the time distribution characteristics based on the active segments and extract the time series features;

[0067] S4. Perform cluster analysis on the user's online behavior based on the user's active time in the time series characteristics to obtain multiple user behavior categories, and classify the recommendation data according to the user behavior categories to obtain preferred recommended content and alternative recommended content;

[0068] S5. Dynamically adjust the preferred recommended content and the alternative recommended content based on the feedback scores of each user behavior category, and perform data recommendation based on the dynamically adjusted preferred recommended content and the alternative recommended content.

[0069] As shown in the above steps S1-S5, with the rapid development of the Internet, the online behavior data of users has shown explosive growth. How to extract valuable information from massive data and provide users with personalized recommendation services has become a hot topic in the industry. In this embodiment, it is necessary to first comprehensively collect the online behavior data of users. These online behavior data cover the user's operation records, browsing records and purchase records at various time nodes, so as to fully understand the user's network behavior. Then, a series of preprocessing tasks need to be performed on the collected online behavior data, including data cleaning, data deduplication, etc., to ensure the accuracy and validity of the data. Then, the preprocessed data is quantitatively mapped to obtain benchmark feature parameters that can reflect the user's behavior characteristics and belong to the same standard. Then, multiple parallel time periods are constructed. The length of the parallel time period is preferably one day, which can be set according to actual needs. Then, Based on the occurrence time of each benchmark feature parameter, these parallel time periods are divided into active segments and blank segments, so that the user's active time period can be grasped more accurately. Then, the time distribution feature analysis of the active segments under each parallel time period is performed to extract key time series features. These features mainly include the user's active duration and active frequency. On this basis, based on these time series features, the user's online behavior is clustered and analyzed to obtain multiple different user behavior categories. Then, based on these user behavior categories, the recommendation data is classified and processed to obtain preferred recommended content and alternative recommended content. Finally, the user feedback scores under each user behavior category are obtained in real time. These feedback scores will serve as the basis for adjusting the recommended content. That is, based on the feedback scores, the preferred recommended content and alternative recommended content can be dynamically adjusted, and a recommendation list can be generated simultaneously to provide recommended content that better meets user needs.

[0070] In a preferred embodiment, the steps of preprocessing the acquired online behavior data of the user and performing quantitative mapping processing on the preprocessed online behavior data to obtain the baseline feature parameters include:

[0071] S101. Obtain online behavior data of all users and perform data cleansing on the online behavior data to remove duplicate and abnormal records in the online behavior data to obtain cleansed data;

[0072] S102: Acquire feature extraction conditions, and perform feature extraction on the cleaned data based on the feature extraction conditions to obtain demand features, wherein the feature extraction conditions include the user's operation records, browsing records, and purchase records at different time points;

[0073] S103. Convert all required features into quantitative values ​​of the same standard through quantitative mapping, and record the quantitative values ​​as reference feature parameters.

[0074] As described in the above steps S101-S103, online behavior data includes user click behavior, browsing history, search history, purchase behavior, etc. After collecting the online behavior data, it needs to be cleaned. The cleaning process mainly includes removing duplicate records and abnormal records in the data. Duplicate records may be generated due to various reasons in the data collection process, and abnormal records may be due to errors generated during data transmission or storage. Through cleaning, a clean and accurate data set can be obtained. Then, it is necessary to determine the conditions for feature extraction, and extract features from the cleaned data according to the feature extraction conditions to obtain the required demand features. Then, all extracted demand features are quantitatively mapped. Quantitative mapping is to convert non-numeric data into numerical data for subsequent data analysis and processing. Finally, the quantitative mapping results of all demand features are recorded as benchmark feature parameters. Its purpose is to convert demand features into data in the same standard format for subsequent data analysis and processing. In this process, the standard function involved in the quantitative mapping process is: C x =(CC min ) / (C max -C min ), where C x represents the quantification result, C represents the demand characteristics to be quantified (demand characteristics include but are not limited to user operation records, browsing records and purchase records at different time points), C min Indicates the minimum value of the quantified demand characteristic, C max Indicates the maximum value of the quantified demand characteristic.

[0075] In a preferred embodiment, the step of obtaining active segments from each parallel time period according to the occurrence time of each benchmark characteristic parameter includes:

[0076] S201, performing backtracking offset based on the current time node as a reference point, and outputting a monitoring period based on the backtracking offset result;

[0077] S202: Divide the monitoring period into multiple equally spaced time periods, and record the multiple equally spaced time periods as parallel time periods;

[0078] S203: Count the number of occurrences and duration of user behaviors in each parallel time period based on the occurrence time nodes of each benchmark characteristic parameter, and record the duration as the user online period. Record the period in which no user behavior occurs as a blank period.

[0079] S204: Filter the duration of the user's online period to obtain active segments and occasional segments, and merge the occasional segments with adjacent blank segments to obtain adjusted blank segments and active segments.

[0080] As described in the above steps S201-S204, when determining the user's active time period, the current time point is first selected as the benchmark point, and then backtracking is performed based on the benchmark point to determine a time range as a monitoring period. Within this monitoring period, the entire monitoring period will be divided into multiple time periods at equal intervals. These time periods are regarded as parallel time periods. The time length of the parallel time period can be one day, one week, etc. Subsequently, based on the occurrence time points of the benchmark feature parameters in each time period, the number of occurrences and duration of user behaviors in each time period are counted, and the time period when such user behaviors occur is recorded as an online time period. Correspondingly, the time period when no user behavior occurs is directly identified and recorded as a blank segment. Finally, the length of these user online time periods is screened accordingly, so that active time periods and occasional time periods can be effectively identified. For occasional time periods, they will be merged with adjacent blank time periods to create an adjusted time period distribution, and this distribution includes active segments and adjusted blank segments.

[0081] In a preferred embodiment, the steps of filtering the length of the user's online time period to obtain active segments and sporadic segments, and merging the sporadic segments with adjacent blank segments to obtain adjusted blank segments and active segments include:

[0082] Step 1: Obtain all user online time periods and record the duration of the user online time period as the parameter to be filtered;

[0083] Step 2: Get the screening threshold and compare it with the parameter to be screened;

[0084] If the parameter to be filtered is greater than or equal to the filtering threshold, the user's online period is marked as an active period;

[0085] If the parameter to be filtered is less than the filtering threshold, the user's online period is marked as an occasional period, and when the occasional period is continuous, it is merged into an active segment; otherwise, it is merged with the adjacent blank segment.

[0086] As described in Step 1-Step 2 above, after the user's online time period is output, the active period and the occasional period will be further determined. First, it is necessary to collect and organize the online periods of all users, and record the length of each user's online period as a parameter to be screened. Next, it is necessary to determine a screening threshold. The screening threshold will be used as a standard to distinguish whether the user's online period is in an active state or an occasional state. The setting of the screening threshold needs to comprehensively consider user behavior characteristics and business needs to ensure the accuracy and effectiveness of data analysis. Then, the screening parameter is compared with the screening threshold. If the value of the screening parameter is greater than or equal to the screening threshold, it means that the length of the user's online period meets the definition of an active segment, and this part of the period is directly marked as an active segment. It is recorded as an active segment. On the contrary, if the value of the parameter to be filtered is less than the filtering threshold, it means that the length of the user's online period is short and belongs to an occasional period. An occasional period usually represents a user's short online behavior, which may be due to non-continuous activities such as browsing and querying information. For continuous time periods marked as occasional periods, they will be merged into a unified active segment, because continuous occasional periods may actually reflect a continuous activity process, and merging them into active segments can more accurately describe the user's behavior pattern. For occasional periods that cannot be merged with adjacent blank segments, they will be further merged with adjacent blank segments to ensure continuity and integrity on the timeline, so that the data analysis results can more truly reflect the user's behavior patterns.

[0087] In a preferred embodiment, the step of performing time distribution feature analysis based on the active segment and extracting time series features includes:

[0088] S301: Obtain active segments in each parallel time period, and calculate the total duration of active segments and the total duration of blank segments in each parallel time period;

[0089] S302: The ratio of the total duration of the active segment to the duration of the parallel period is used as the user's activity rate in the parallel period, and the activity rates of all parallel periods are counted to obtain an activity rate time series;

[0090] S303: Count the number of active users in each parallel period to obtain an active frequency time series;

[0091] S304: Use the activity rate time series and the activity frequency time series as the time series features of the user in the time period.

[0092] As described in the above steps S301-S304, when analyzing the time distribution characteristics of active segments, first collect the active segments of users in each parallel time period, and count the total duration of active segments in each time period, and then calculate the proportion of the total duration of active segments in each parallel time period, and then output the activity rate. Then, summarize and count the activity rates of all parallel time periods to form a time series data set of activity rates. Furthermore, it is necessary to separately count the number of active times of users in each parallel time period, which can allow us to understand the frequency distribution of users' activity in each time period. Based on this, we can construct a time series of active frequencies to understand the concentration and fluctuation of users' activity. Finally, the active rate time series and the active frequency time series are integrated as the time series characteristics of users in the time period, providing corresponding data support for the subsequent analysis of user behavior categories.

[0093] In a preferred embodiment, the step of performing cluster analysis on the user's online behavior based on the user's active duration in the time series feature to obtain multiple user behavior categories includes:

[0094] S401: Obtain the active duration of each active segment in each parallel time period, and the proportion of the active duration of each active segment in the corresponding parallel time period, and record the proportion as a condition parameter;

[0095] S402: Obtain a clustering interval, wherein a plurality of clustering intervals are provided, and each clustering interval corresponds to a user behavior category;

[0096] S403: Compare the condition parameters with the clustering intervals one by one, and output the user behavior category under each active segment based on the comparison results.

[0097] As described in steps S401-S403 above, when performing cluster analysis on the user's online behavior, first collect and organize the user's active segment duration in different time periods, that is, the active duration, and the proportion of this duration in the entire time period, and record them as conditional parameters. Then, determine the specific interval for cluster analysis. Specifically, multiple clustering intervals can be set based on business needs, user behavior characteristics, historical data and other factors. Each clustering interval corresponds to a specific user behavior category, such as "high-frequency active users", "low-frequency active users", and "occasional users". The clustering interval is pre-set and contains multiple different clusters, each of which corresponds to a user behavior category. Then, the conditional parameters are compared one by one with these clustering intervals. Through this comparison, the user group to which each active segment belongs can be determined. Finally, based on the comparison results, the user behavior category corresponding to each active segment can be output. Through the above cluster analysis, the user's online behavior can be classified according to its active duration and proportion, so as to more accurately understand the user's behavior pattern.

[0098] In a preferred embodiment, the step of classifying the recommendation data according to the user behavior category to obtain the preferred recommendation content and the alternative recommendation content includes:

[0099] S404: Obtain the browsing time and purchase history of each recommended content by the user under each user behavior category, and then mark the browsing time and purchase history as the first evaluation parameter and the second evaluation parameter, respectively;

[0100] S405: Obtain an evaluation function, input the first evaluation parameter and the second evaluation parameter into the evaluation function, and mark the output result of the evaluation function as an evaluation score;

[0101] S406: Obtain an evaluation threshold, and compare the evaluation threshold with the evaluation score;

[0102] If the evaluation score is greater than or equal to the evaluation threshold, the corresponding content to be recommended is marked as preferred recommended content;

[0103] If the evaluation score is less than the evaluation threshold, the corresponding content to be recommended is marked as candidate recommendation content.

[0104] As described in steps S404-S406 above, after the user behavior categories are determined, the browsing time and purchase history of recommended content for different users under different behavior categories are first collected. In this embodiment, the user's browsing time and purchase history are used as evaluation indicators, and are defined as a first evaluation parameter and a second evaluation parameter, respectively. Then, an evaluation function is introduced, and the first evaluation parameter and the second evaluation parameter are used as inputs. Through calculation by the evaluation function, an evaluation score is output;

[0105] Among them, the expression of the evaluation function is: In the formula, r represents the evaluation score, α and β represent the weight coefficients of the first evaluation parameter and the second evaluation parameter, respectively, which are used to adjust the relative importance of browsing time and purchase records in the evaluation, n and m represent the number of the first evaluation parameter and the second evaluation parameter, respectively, and x i and y i represent the first evaluation parameter and the second evaluation parameter respectively;

[0106] Then set an evaluation threshold. The evaluation threshold is a standard for distinguishing preferred recommended content from alternative recommended content. In actual applications, each evaluation score is compared with the evaluation threshold, and the recommended content is classified according to the comparison results. If the evaluation score of the recommended content is greater than or equal to the evaluation threshold, then the recommended content is considered to be high-quality, because the corresponding user browsing and purchase volume of the recommended content is large, and the recommended content will be marked as preferred recommended content. On the contrary, if the evaluation score is less than the evaluation threshold, then the recommended content is considered to be relatively poor, and the recommended content will be marked as alternative recommended content for subsequent backup, so as to ensure the accuracy of the recommendation and meet the personalized needs of users.

[0107] In a preferred embodiment, the steps of dynamically adjusting the preferred recommended content and the alternative recommended content based on the feedback scores of each user behavior category, and performing data recommendation based on the dynamically adjusted preferred recommended content and the alternative recommended content include:

[0108] S501: Construct a feedback statistical period and collect feedback scores of users on recommended content under various user behavior categories within the feedback statistical period;

[0109] S502: After the feedback statistics period ends, a standard function is obtained, the feedback score is input into the standard function, and the output value of the standard function is recorded as the recommendation optimization score;

[0110] S503: Re-rank the preferred recommended content and the alternative recommended content according to the recommendation optimization score, and perform data recommendation based on the ranking results of the preferred recommended content and the alternative recommended content.

[0111] As described in steps S501-S503 above, in order to accurately capture and utilize user feedback scores so as to more effectively adjust recommended content and generate recommendation lists, it is first necessary to establish a feedback statistical period. The length of the feedback statistical period needs to be set according to specific needs so as to collect representative user feedback data. During this period, user feedback scores on recommended content under various user behavior categories will be monitored and collected. This includes users' click-through rate, browsing time, likes, favorites, shares, and other interactive behaviors on recommended content, as well as their direct ratings or comments. At the end of the feedback statistical period, a standard function is introduced to convert the feedback scores into recommendation optimization scores.

[0112] Among them, the expression of the standard function is:

[0113] In the formula, U represents the recommendation optimization score, η represents the quantitative coefficient, T represents the length of the feedback statistical period, a represents the number of feedback scores, and f k1 and f k2 All indicate feedback ratings;

[0114] Based on the above formula, the feedback trend and overall feedback of users on the recommended content during the feedback statistical period are summarized and processed, so as to output the corresponding recommendation optimization score. Finally, the preferred recommended content and alternative recommended content are re-ranked according to these recommendation optimization scores to ensure the quality of the recommended content. The sorted recommended content is output as a recommendation list, and the data recommendation process is performed based on the recommendation list. The recommendation list is not only a collection of preferred recommended content, but also includes alternative recommended content, so as to provide other options when the preferred recommended content does not meet the user's needs.

[0115] During this process, user feedback can be periodically fed back according to the length of the feedback statistical period. The beginning and end of the feedback statistical period can be connected uninterruptedly. The setting of the feedback statistical period can ensure the adequacy of the feedback data, thereby ensuring the accuracy of the recommendation optimization score output. Correspondingly, it can also achieve periodic dynamic adjustment of the preferred recommended content and the alternative recommended content to ensure the timeliness and accuracy of the recommended content.

[0116] See also Figure 2 , a data recommendation system based on time series features, including:

[0117] The data acquisition module is used to pre-process the acquired online behavior data of users and perform quantitative mapping on the pre-processed online behavior data to obtain benchmark feature parameters;

[0118] The time period division module is used to obtain active segments from each parallel time period according to the occurrence time of each benchmark characteristic parameter;

[0119] Feature analysis module: The feature analysis module is used to analyze the time distribution characteristics based on the active segments and extract time series features;

[0120] Cluster analysis module: The cluster analysis module is used to perform cluster analysis on users' online behaviors based on the user's active time in the time series characteristics, obtain multiple user behavior categories, and classify the recommended data according to the user behavior categories to obtain preferred recommended content and alternative recommended content;

[0121] The dynamic adjustment module is used to dynamically adjust the preferred recommended content and the alternative recommended content according to the feedback scores of each user behavior category, and to make data recommendations based on the dynamically adjusted preferred recommended content and the alternative recommended content.

[0122] As mentioned above, the system includes a data acquisition module, a time period division module, a feature analysis module, a cluster analysis module and a dynamic adjustment module. The data acquisition module is responsible for collecting users' network behavior data. These data cover multiple aspects such as users' operation records, browsing history and purchase records at various time nodes, thereby providing rich original data for subsequent data processing and analysis. At the same time, the collected network behavior data will be preliminarily cleaned and sorted to remove invalid and duplicate data and improve data quality. In addition, the processed data will be quantitatively mapped and converted into benchmark feature parameters that can be quantitatively analyzed and processed, providing corresponding data support for subsequent time period division and feature analysis. The time period division module is based on the occurrence time of these benchmark feature parameters to build multiple parallel time periods, and based on this, each time period is divided into active segments and blank segments, so as to further understand the user's behavioral habits. The feature analysis module will conduct in-depth time distribution feature analysis on these divided active segments, and extract key time series features from them. Time series features include but are not limited to users' active duration and active frequency, so as to understand users' behavioral patterns and preferences. The clustering analysis module will cluster the user's behavior based on these time series features, thereby forming multiple different user behavior categories. The dynamic adjustment module is responsible for collecting users' feedback scores on recommended content, and based on this, dynamically adjusts the recommended content to ensure the accuracy and timeliness of the recommended content. It will also simultaneously generate a recommendation list to provide users with real-time personalized recommendations.

[0123] See also Figure 3 , an electronic device, the electronic device comprising:

[0124] at least one processor;

[0125] and a memory communicatively coupled to the at least one processor;

[0126] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can execute the above-mentioned data recommendation method based on time series features.

[0127] The processor of the electronic device mentioned above may be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The memory may include volatile memory (such as random access memory RAM) and / or non-volatile memory (such as read-only memory ROM, flash memory, etc.), and also includes an operator, which may be an arithmetic logic unit (ALU), which is responsible for performing various arithmetic and logical operations, such as addition, subtraction, multiplication, division, bit operations, etc., to ensure that the electronic device can accurately complete various computing tasks when executing the data recommendation method based on time series features.

[0128] In an electronic device, a processor and a memory are communicatively connected via a bus or the like, allowing the processor to read and execute a computer program stored in the memory, thereby implementing a data recommendation method based on time series features. Furthermore, the electronic device may also include other components communicatively connected to the processor, such as input / output devices, network interface cards, and display devices, to support user interaction and data exchange with external devices.

[0129] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0130] The foregoing is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained herein shall, unless otherwise specified or limited, be implemented in accordance with conventional means in the art.

Claims

1. A data recommendation method based on time series features, characterized in that: include: Preprocessing the acquired online behavior data of the user, and performing quantitative mapping processing on the preprocessed online behavior data to obtain benchmark feature parameters; Obtaining active segments from each parallel time period according to the occurrence time of each of the benchmark characteristic parameters; Performing time distribution feature analysis based on the active segments to extract time series features; Perform cluster analysis on the online behavior of users based on the user active time in the time series characteristics to obtain multiple user behavior categories, and classify the recommended data according to the user behavior categories to obtain preferred recommended content and alternative recommended content; including: obtaining the active time of each active segment in each of the parallel time periods, and the proportion of the active time of each active segment in the corresponding parallel time period, and recording the proportion as a condition parameter; obtaining a clustering interval, wherein there are multiple clustering intervals, and each clustering interval corresponds to a user behavior category; comparing the condition parameters with the clustering intervals one by one, and outputting the user behavior category under each active segment based on the comparison result; wherein, multiple clustering intervals are set according to business needs, user behavior characteristics and historical data factors, and each clustering interval is a clustering interval. Each cluster interval corresponds to a specific user behavior category, and the user behavior categories include: high-frequency active users, low-frequency active users, and occasional users; obtain the user's browsing time and purchase record for each recommended content under each user behavior category, and then calibrate the browsing time and purchase record as a first evaluation parameter and a second evaluation parameter respectively; obtain an evaluation function, and input the first evaluation parameter and the second evaluation parameter into the evaluation function, and calibrate the output result of the evaluation function as an evaluation score; obtain an evaluation threshold, and compare the evaluation threshold with the evaluation score; if the evaluation score is greater than or equal to the evaluation threshold, mark the corresponding content to be recommended as preferred recommended content; if the evaluation score is less than the evaluation threshold, mark the corresponding content to be recommended as alternative recommended content; the expression of the evaluation function is: In the formula, r represents the evaluation score, α and β represent the weight coefficients of the first evaluation parameter and the second evaluation parameter, respectively, which are used to adjust the relative importance of browsing time and purchase records in the evaluation, n and m represent the number of the first evaluation parameter and the second evaluation parameter, respectively, and x i and y i represent the first evaluation parameter and the second evaluation parameter respectively; The preferred recommended content and the alternative recommended content are dynamically adjusted according to the feedback scores of each of the user behavior categories, and data recommendation is performed based on the dynamically adjusted preferred recommended content and the alternative recommended content, including constructing a feedback statistical period and collecting feedback scores of users on the recommended content under each of the user behavior categories within the feedback statistical period; after the execution of the feedback statistical period ends, a standard function is obtained, and the feedback score is input into the standard function, and the output value of the standard function is recorded as the recommendation optimization score; the preferred recommended content and the alternative recommended content are re-sorted according to the recommendation optimization score, and data recommendation is performed based on the sorting results of the preferred recommended content and the alternative recommended content.

2. The data recommendation method based on time series features according to claim 1, characterized in that: The step of preprocessing the acquired online behavior data of the user and performing quantitative mapping processing on the preprocessed online behavior data to obtain the benchmark feature parameters includes: Acquire online behavior data of all users, and perform data cleaning on the online behavior data to remove duplicate and abnormal records in the online behavior data to obtain cleaned data; Acquiring feature extraction conditions, and performing feature extraction on the cleaned data according to the feature extraction conditions to obtain demand features, wherein the feature extraction conditions include the user's operation records, browsing records, and purchase records at different time points; All the required characteristics are converted into quantitative values ​​of the same standard through quantitative mapping, and the quantitative values ​​are recorded as the reference characteristic parameters.

3. The data recommendation method based on time series features according to claim 2, characterized in that: The step of obtaining active segments from each parallel time period according to the occurrence time of each of the reference characteristic parameters includes: Perform backtracking offset based on the current time node and output the monitoring period based on the backtracking offset result; The monitoring period is divided into a plurality of equally spaced time periods, and the plurality of equally spaced time periods are recorded as parallel time periods; According to the occurrence time nodes of each of the benchmark characteristic parameters, the number of occurrences and duration of the user behavior in each of the parallel time periods are counted, and the duration is recorded as the user online period, and the period in which no user behavior occurs is directly recorded as a blank period; The length of the online time period of the user is screened to obtain active segments and sporadic segments, and the sporadic segments are merged with adjacent blank segments to obtain adjusted blank segments and active segments.

4. The data recommendation method based on time series features according to claim 3, characterized in that: The step of screening the length of the user's online time period to obtain active segments and sporadic segments, and merging the sporadic segments with adjacent blank segments to obtain adjusted blank segments and active segments includes: Obtaining all online time periods of the users, and recording the duration of the online time periods of the users as parameters to be screened; Obtaining a screening threshold, and comparing the screening threshold with the parameter to be screened; If the parameter to be screened is greater than or equal to the screening threshold, marking the user's online period as an active period; If the parameter to be screened is less than the screening threshold, the user's online period is marked as an occasional period, and when the occasional period is continuous, it is merged into an active segment; otherwise, it is merged with an adjacent blank segment.

5. The data recommendation method based on time series features according to claim 4, characterized in that: The step of performing time distribution feature analysis based on the active segment to extract time series features includes: Obtaining the active segments in each of the parallel time periods, and calculating the total duration of the active segments and the total duration of the blank segments in each of the parallel time periods; The ratio of the total duration of the active segment to the duration of the parallel period is used as the user's activity rate in the parallel period, and the activity rates of all parallel periods are counted to obtain an activity rate time series; Counting the number of active users in each of the parallel time periods to obtain an active frequency time series; The activity rate time series and the activity frequency time series are used as the time series features of the user in the time period.

6. A data recommendation method based on time series features according to claim 5, applied to a data recommendation system based on time series features, characterized in that: include: A data acquisition module, configured to preprocess the acquired online behavior data of the user and perform quantitative mapping on the preprocessed online behavior data to obtain benchmark feature parameters; a time period division module, the time period division module being configured to obtain active segments from each parallel time period according to the occurrence time of each of the reference characteristic parameters; A feature analysis module, configured to perform time distribution feature analysis based on the active segment and extract time series features; A cluster analysis module, configured to perform cluster analysis on the user's online behavior based on the user's active duration in the time series feature to obtain multiple user behavior categories, and to classify the recommendation data according to the user behavior categories to obtain preferred recommended content and alternative recommended content; A dynamic adjustment module is used to dynamically adjust the preferred recommended content and the alternative recommended content according to the feedback scores of each of the user behavior categories, and to perform data recommendation based on the dynamically adjusted preferred recommended content and the alternative recommended content.

7. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the data recommendation method based on time series features described in any one of claims 1 to 5.

Images