News recommendation method and device, electronic equipment and computer readable storage medium

Abstract

The present disclosure provides a news recommendation method and device, electronic equipment and computer readable storage medium, and relates to the technical field of big data. The method comprises the following steps: constructing a keyword list of each news to be recommended; obtaining a keyword frequent item combination generated by a target user when browsing news in the past; scoring each news to be recommended according to the keywords in the keyword list and the keyword frequent item combination, to obtain a news score of each news to be recommended; and determining target news to be recommended to the target user from all the news to be recommended according to the news score of each news to be recommended. The present disclosure can not only recommend news to users with sufficient information, but also recommend news to users with less information, which is conducive to improving the accuracy and recommendation effect of news recommendation.

Description

Technical Field

[0001] This disclosure relates to the field of big data technology, and in particular to a news recommendation method, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] With the rapid development of internet technology, the advantages of news recommendation have become increasingly apparent. News recommendations allow people to easily stay informed about work and life events. However, current news recommendations are based on limited past browsing data (such as browsing time and page views), failing to consider users' evolving needs. This results in inaccurate news recommendations and ultimately, poor recommendation performance.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] This disclosure provides a news recommendation method, apparatus, electronic device, and computer-readable storage medium that improves the accuracy of news recommendations to at least a certain extent.

[0005] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0006] According to one aspect of this disclosure, a news recommendation method is provided, the news recommendation method comprising:

[0007] Build a keyword list for each news item to be recommended;

[0008] Obtain frequent keyword combinations generated from the target user's past news browsing history;

[0009] Based on the keywords in the keyword list and the frequent combinations of the keywords, each of the above-mentioned news items to be recommended is scored to obtain a news score for each of the above-mentioned news items to be recommended.

[0010] Based on the news rating of each of the aforementioned news items to be recommended, target news items are identified from all of the aforementioned news items to be recommended and recommended to the aforementioned target users.

[0011] Preferably, the steps of constructing the keyword list for each news item to be recommended include: performing word segmentation on the text of each news item to be recommended to obtain multiple word segments; setting a weight value for each word segment using a term frequency inverse document frequency algorithm; using word segments with weight values ​​greater than a preset value as keywords for each news item to be recommended; and constructing the keyword list based on the keywords for each news item to be recommended.

[0012] Preferably, the step of scoring each of the aforementioned news items to be recommended based on the keywords in the keyword list and the frequent combinations of the keywords to obtain a news score for each of the aforementioned news items to be recommended includes: detecting whether the keywords in the keyword list correspond to attribute information; if the keywords in the keyword list correspond to attribute information, obtaining the attribute information of the keywords in the keyword list, and determining the keyword score of the keywords based on the attribute information; determining the keyword combination score of the frequent combinations of the keywords based on the keyword scores of the keywords in the keyword list; determining a first comprehensive score for each of the aforementioned news items to be recommended based on the sum of the keyword scores, and determining a second comprehensive score for each of the aforementioned news items to be recommended based on the sum of the keyword combination scores; and determining the news score for each of the aforementioned news items to be recommended based on the first comprehensive score and the second comprehensive score.

[0013] Preferably, the steps of obtaining attribute information of keywords in the keyword list and determining keyword scores based on the attribute information include: dividing the keywords in the keyword list into a first part of keywords and a second part of keywords; obtaining attribute information of the first part of keywords; wherein the attribute information includes time data generated from browsing news, user information, and browsing data; filtering the minimum value in the browsing data to obtain intermediate data; sorting the values ​​in the intermediate data to obtain sorted data, and calculating a set quantile in the sorted data; adjusting the values ​​in the sorted data that are greater than or equal to the set quantile to the set quantile to obtain preprocessed data; standardizing the preprocessed data to obtain target data; determining a first score for the first part of keywords based on the time data and the target data; determining a second score for the second part of keywords based on a model-based collaborative filtering recommendation algorithm, the time data, the user information, the first part of keywords, and the first score; and determining the keyword score based on the first score and the second score.

[0014] Preferably, the step of determining the keyword combination score of the frequent keyword combination based on the keyword scores of the keywords in the keyword list includes: when the keyword list contains the frequent keyword combination, the keyword scores of the keywords included in the frequent keyword combination are summed to obtain the keyword combination score of the frequent keyword combination.

[0015] Preferably, the step of determining the target news to recommend to the target user from all the news to be recommended based on the news rating of each of the news to be recommended includes: determining a first candidate news from all the news to be recommended based on the ranking of the first comprehensive score in the news rating; determining a second candidate news from all the news to be recommended based on the ranking of the second comprehensive score in the news rating; and determining the first candidate news and the second candidate news as the target news to recommend to the target user.

[0016] Preferably, the step of scoring each of the aforementioned news items to be recommended based on the keywords in the keyword list and the frequent keyword combinations to obtain a news score for each of the aforementioned news items to be recommended includes: detecting whether the keywords in the keyword list correspond to attribute information; if the keywords in the keyword list do not correspond to attribute information, and the keyword list contains the frequent keyword combinations, determining the keyword combination score of the frequent keyword combination based on the number of keywords contained in the frequent keyword combination; or, setting a preset score for the frequent keyword combination to obtain the keyword combination score; and determining the news score for each of the aforementioned news items to be recommended based on the sum of the keyword combination scores.

[0017] Preferably, the step of obtaining the frequent keyword combinations generated by the target user's previous browsing of news includes: using a frequent itemset mining algorithm to obtain the frequent keyword combinations generated by the target user's previous browsing of news.

[0018] Preferably, after the step of determining the target news to be recommended to the target user from all the news to be recommended based on the news rating of each of the news to be recommended, the news recommendation method further includes: obtaining the target user's preferred time for browsing news in the past; dividing the day into multiple time periods, determining the time period in which the preferred time period is located, and obtaining the recommended time period; and recommending the target news to the target user based on the recommended time period.

[0019] According to another aspect of this disclosure, a news recommendation device is provided, the news recommendation device comprising:

[0020] The list building module is used to build a list of keywords for each news item to be recommended;

[0021] The combination acquisition module is used to acquire frequent keyword combinations generated from the target user's past browsing of news;

[0022] The news scoring module is used to score each of the above-mentioned news items to be recommended based on the keywords in the keyword list and the combination of the most frequent keywords, and to obtain the news score for each of the above-mentioned news items to be recommended.

[0023] The news selection module is used to determine the target news to recommend to the target user from all the aforementioned news items based on the news rating of each of the aforementioned news items to be recommended.

[0024] According to another aspect of this disclosure, an electronic device is provided, the electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the news recommendation method as described in the above embodiments.

[0025] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the news recommendation method as described in the above embodiments.

[0026] The news recommendation method, apparatus, electronic device, and computer-readable storage medium provided in the embodiments of this disclosure have the following technical effects:

[0027] This disclosure employs a technical solution that constructs a keyword list for each news item to be recommended, obtains frequent keyword combinations generated by the target user's past news browsing, scores each news item based on the keywords and frequent keyword combinations in the keyword list, and determines the target news to recommend to the target user from all the news items to be recommended. This solution can recommend news to users with ample information as well as those with limited information. Regardless of whether the news recommendation is for users with ample or limited information, the information used for news scoring comes from both the keywords contained in each news item to be recommended and information related to the user's past habits, i.e., frequent keyword combinations. By scoring each news item using keywords and frequent keyword combinations, the resulting news score is not only closely related to the user's past search keywords but also to the attribute information of each keyword in each news item to be recommended, which helps improve the accuracy and effectiveness of news recommendations.

[0028] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0029] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0030] Figure 1 A flowchart illustrating the news recommendation method provided in this embodiment of the present disclosure is shown.

[0031] Figure 2 An exemplary flowchart of S110 in the news recommendation method provided in this embodiment is shown;

[0032] Figure 3 An exemplary flowchart of S130 in the news recommendation method provided in this embodiment is shown;

[0033] Figure 4 Another exemplary flowchart of S130 in the news recommendation method provided in this disclosure embodiment is shown;

[0034] Figure 5 A schematic diagram of the structure of a news recommendation device to which an embodiment of the present disclosure can be applied is shown;

[0035] Figure 6 A schematic diagram of the structure of an electronic device in an exemplary embodiment of the present disclosure is shown. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this disclosure clearer, the embodiments of this disclosure will be described in further detail below with reference to the accompanying drawings.

[0037] In the following description, when referring to the accompanying drawings, the same numbers in different drawings denote the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0038] It should be noted that the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this disclosure, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0039] News recommendations allow people to stay informed about work and life events. However, current news recommendations are based on users' historical browsing habits (such as browsing time and page views), failing to consider new user needs. This results in inaccurate news recommendations that don't meet users' actual requirements. Existing ALS (Alternating Least Squares) recommendation algorithms are ineffective against cold-start problems. ALS requires users to have a certain amount of accumulated information; for users with insufficient information, the recommendations will be inaccurate, leading to poor results. Furthermore, while existing algorithms can provide recommendations to a thousand users, they don't consider the cyclical changes in user needs over time, resulting in recommendations that don't address new user demands and ultimately contributing to poor recommendation performance.

[0040] Based on the problems existing in the aforementioned related technologies, this disclosure provides a news recommendation method, apparatus, electronic device, and computer-readable storage medium. When users lack sufficient valid historical information, the traditional ALS recommendation algorithm performs poorly. This disclosure employs a frequent itemset approach for news recommendation for new users or users with insufficient information. For users with sufficient information, frequent itemset recommendation can be combined with ALS recommendation. First, the ALS algorithm calculates the score of frequent itemsets; then, the top-scoring frequent itemsets are selected; finally, frequent itemsets containing the specified tag are identified, and these frequent itemsets are used for news recommendation. This algorithm supplements and improves the ALS recommendation algorithm. The ALS recommendation algorithm is also known as the ALS collaborative filtering algorithm; frequent itemsets are mined using the Frequent Pattern Growth (FP-growth) algorithm, also known as the association rule algorithm. In this disclosure, "tag" refers to a keyword, and a frequent itemset refers to a frequent itemset containing at least one keyword.

[0041] The news recommendation method disclosed herein is used for personalized news recommendations. It combines the time patterns of each user's news viewing to recommend different news at different times, adding time-based variation to the basic personalized recommendation approach. Furthermore, news recommendations differ from traditional product recommendations. News content is text data, which is unstructured, while traditional product recommendations are based on structured data. To address this characteristic, this disclosure combines the FPGrowth association rule algorithm with the ALS collaborative filtering algorithm to improve the accuracy and effectiveness of news recommendations.

[0042] The following are embodiments of the news recommendation method provided in this disclosure.

[0043] Figure 1 A flowchart illustrating the news recommendation method provided in an embodiment of this disclosure is shown. Figure 1 As shown, an embodiment of this disclosure provides a news recommendation method, which includes the following solutions:

[0044] S110: Construct a keyword list for each news item to be recommended.

[0045] In one exemplary embodiment, multiple news items to be recommended are obtained. After obtaining the news items to be recommended, a corresponding keyword list is constructed for each news item to be recommended. The keyword list includes multiple keywords, and the keywords in the keyword list are words in the text of the corresponding news item to be recommended.

[0046] S120: Obtain frequent keyword combinations generated from the target user's past browsing of news.

[0047] Frequent keyword combinations include at least one preferred keyword. Preferred keywords can be understood as search terms that users have frequently used in their past news searches, also known as historical search keywords. For example, if a user searches for news A using the search term "track and field," then "track and field" is a frequent keyword combination. As another example, if a user searches for news B using the search terms "technology," "robot," and "robotic arm," then "technology," "robot," and "robotic arm" constitute a frequent keyword combination.

[0048] Optionally, S120 includes: using a frequent itemset mining algorithm to obtain the frequent item combinations of the keywords generated by the target user's previous browsing of news.

[0049] Frequent itemset mining algorithms refer to the FP-growth algorithm mentioned above. The target user can be understood as the object of news recommendations. The FP-growth algorithm is used to mine frequently used historical search keywords by the target user during past news browsing, thus obtaining frequent item sets of keywords. The obtained frequent item sets of keywords may be one or multiple.

[0050] S130: Based on the keywords in the keyword list and the frequent combinations of the keywords, score each of the above-mentioned news items to be recommended, and obtain the news score for each of the above-mentioned news items to be recommended.

[0051] After obtaining the frequent keyword combinations corresponding to the target user, the inclusion relationship between the keyword list of each news item to be recommended and each frequent keyword combination is determined. That is, if at least one preferred keyword in each frequent keyword combination is included in the keyword list, the inclusion relationship is that the keyword list includes the frequent keyword combination. Then, the preferred keywords included in the keyword list are used to score the frequent keyword combination, resulting in a keyword combination score. Specifically, if at least one preferred keyword in a frequent keyword combination is included in the keyword list, it can also be understood that the frequent keyword combination contains some keywords from the keyword list; that is, the frequent keyword combination contains the same preferred keywords as keywords in the keyword list. For example, if the frequent keyword combination includes the preferred keywords "football" and "World Cup," and the keyword list also contains keywords with the same meaning, then the frequent keyword combination contains some keywords from the keyword list.

[0052] Based on the scoring method for frequent keyword combinations described above, the inclusion relationship between the keyword list of each news item to be recommended and the various frequent keyword combinations corresponding to the target user is used to score all frequent keyword combinations for the target user. Specifically, for the same frequent keyword combination, the more preferred keywords from that combination are included in the keyword list, the higher the keyword combination score for that combination.

[0053] Scoring a frequent keyword combination using preferred keywords included in the keyword list includes at least two scenarios.

[0054] In one scenario, news recommendations can be made to users with sufficient information. This information includes relevant data from the news articles the user has previously viewed, such as keywords in the recommended news articles, keyword attribute information, and frequent keyword combinations relevant to the target user. The attribute information includes time data generated from browsing the recommended news articles, user information, browsing data, news type identifier (ID), keywords, and user biometric information. The time data includes the browsing date and the time period during which the news was viewed, represented by a time ID, such as an hour ID. User information is represented by a user ID. Browsing data can be understood as business attribute information, including the number of views in the last set number of days, the average daily number of views in the last set number of days, the average page dwell time in the last set number of days, the number of views on a statistical day, and the average page dwell time on a statistical day (e.g., a set number of days of 30 days). User biometric information includes gender, age, marital status, and education level.

[0055] After constructing a keyword list for each news item to be recommended, the attribute information of the keywords in the corresponding keyword list is statistically analyzed based on the historical browsing information of each news item. This attribute information is generated after the news item containing keywords from the keyword list has been viewed. The attribute information reflects the user's preference for the keywords corresponding to those attributes. After obtaining the attribute information of each keyword in the keyword list for each news item to be recommended, a score is assigned to each keyword based on this attribute information. Based on this keyword scoring method, all keywords in the keyword list for each news item to be recommended are scored to obtain a keyword score for each keyword in the keyword list.

[0056] After obtaining the keyword score for each keyword in the keyword list, the keyword combination is scored based on the keyword scores of the keywords in the keyword list included in the keyword combination, thus obtaining the keyword combination score for the keyword combination.

[0057] For each news item to be recommended, obtain the keyword score of the keywords in its corresponding keyword list, and the keyword combination score of all frequent keyword combinations for the target user. Then, score each news item by combining the keyword scores of all keywords and the keyword combination scores of all frequent keyword combinations under each news item to be recommended, and obtain the news score of each news item to be recommended.

[0058] For each news item to be recommended, the news score can be determined by summing the keyword scores of all keywords and the keyword combination scores of all frequent keyword combinations, or by scoring the same news item twice using the keyword scores of all keywords and the keyword combination scores of all frequent keyword combinations, resulting in two news scores: one based on keywords and the other based on frequent keyword combinations.

[0059] In another scenario, news recommendations can be made to users with limited information. The relevant information for these users, including keywords from the news articles to be recommended and the news articles to be recommended to the target user, can be used to inform them of their past browsing history.

[0060] After constructing a keyword list for each news item to be recommended, and identifying frequent keyword combinations that contain keywords from other frequent keyword combinations, a keyword combination score is assigned based on the number of keywords in the keyword list contained within that frequent keyword combination. For each news item to be recommended, after obtaining the keyword combination scores for all frequent keyword combinations, a news score is obtained by applying these scores to each news item. One type of news score obtained through this method is based on frequent keyword combinations.

[0061] S140: Based on the news rating of each of the above-mentioned news items to be recommended, determine the target news items to be recommended to the above-mentioned target users from all the above-mentioned news items to be recommended.

[0062] After obtaining the news rating for each news item to be recommended, target news items are selected from all the news items to be recommended to the target user based on their news ratings, and then the recommendations are made. Specifically, if each news item has two news ratings, at least two target news items will be recommended to the target user; if each news item has only one news rating, at least one target news item will be recommended to the target user.

[0063] This embodiment, based on the aforementioned technical solution, employs a method of constructing a keyword list for each news item to be recommended, obtaining frequent keyword combinations generated by the target user's past news browsing, and scoring each news item to be recommended based on the keywords and frequent keyword combinations in the keyword list, thus obtaining a news score for each news item. Based on the news scores of each news item to be recommended, the target news to be recommended to the target user is determined from all the news items to be recommended. This technical solution can recommend news not only to users with ample information but also to users with limited information. Regardless of whether the news recommendation is for users with ample or limited information, the information used for news scoring comes from both the keywords contained in each news item to be recommended and information related to the user's past habits, i.e., frequent keyword combinations. By scoring each news item to be recommended using keywords and frequent keyword combinations, the resulting news score is not only closely related to the user's past search keywords but also to the attribute information of each keyword in each news item to be recommended, which helps improve the accuracy and effectiveness of news recommendations.

[0064] like Figure 2 As shown, Figure 2This illustration shows an exemplary flowchart of step S110 in the news recommendation method provided in this embodiment. Optionally, based on the above method embodiment, step S110 includes the following:

[0065] S111: Perform word segmentation on the text of each of the above-mentioned news items to be recommended, resulting in multiple word segments;

[0066] S112: Use the inverse document frequency algorithm to set a weight value for each of the above word segments;

[0067] S113: Use the word segments with weight values ​​greater than the preset value as keywords for each of the above-mentioned news articles to be recommended;

[0068] S114: Construct the keyword list based on the keywords of each of the above-mentioned news items to be recommended.

[0069] In an exemplary embodiment, the text of the news to be recommended includes at least one of the following: title, summary, and body. Each news article's text is segmented into multiple words. Then, a term frequency-inverse document frequency (TF-IDF) algorithm is used to calculate the weight of each word segment, resulting in a weight value for each word. After obtaining the weight value of each word segment, it is compared with a preset value. If a word segment has a weight value greater than the preset value, it indicates that the word segment with the higher weight value is more important in the corresponding news article, and this word segment can reflect part of the main content of the news article. Therefore, the word segment with the higher weight value is determined as a keyword for the news article. Based on the above keyword selection method, the keywords for each news article can be obtained, and all keywords for each news article can be stored in a list to obtain a keyword list for each news article.

[0070] For example, such as Figure 3 As shown, Figure 3 This illustration shows an exemplary flowchart of step S130 in the news recommendation method provided in this embodiment. Optionally, based on the above method embodiment, step S130 includes the following:

[0071] S1311: Check whether the keywords in the above keyword list have corresponding attribute information;

[0072] S1312: If there is attribute information corresponding to the keywords in the keyword list, obtain the attribute information of the keywords in the keyword list and determine the keyword score of the keywords based on the attribute information.

[0073] S1313: Based on the keyword scores of the keywords in the keyword list above, determine the keyword combination score of the frequent keyword combination above.

[0074] S1314: Determine the first comprehensive score for each of the above-mentioned news items to be recommended based on the sum of the scores of the above-mentioned keywords, and determine the second comprehensive score for each of the above-mentioned news items to be recommended based on the sum of the scores of the above-mentioned keyword combinations.

[0075] S1315: Based on the first comprehensive score and the second comprehensive score mentioned above, determine the news score for each of the above-mentioned news items to be recommended.

[0076] In one exemplary embodiment, a scoring method for calculating the keyword combination score of frequent keyword combinations can be obtained by detecting whether the keywords in the keyword list correspond to attribute information. If the keywords in the keyword list correspond to attribute information, it means that sufficient data that can be used to score the frequent keyword combination can be obtained through relevant information related to the news previously viewed by the user. This data includes the keywords in each of the recommended news articles, the attribute information of the keywords, and the frequent keyword combinations corresponding to the target user. This is a case of recommending news to users with sufficient information. Therefore, when the keywords in the keyword list correspond to attribute information, the attribute information of the keywords in the keyword list is obtained, and the keyword score of the keyword is calculated based on the attribute information. The attribute information reflects the user's degree of liking for the keyword corresponding to the attribute information. The degree of liking can be represented by a specific numerical value. The higher the degree of liking, the higher the keyword score.

[0077] After calculating the keyword scores of all the keywords in the keyword list, the keyword frequent item combination is scored based on the keyword scores of the keywords contained in it. The more keywords a keyword frequent item combination contains, the higher its keyword combination score will be.

[0078] Based on the scenario of recommending news to users with ample information, the scoring for the same news item to be recommended includes two types of scores: Keyword scoring, where the keyword scores of all keywords included in each news item are summed, and this sum is used as the first comprehensive score for each news item. Keyword combination scoring, where the keyword combination scores of all frequent keyword combinations included in each news item are summed, and this sum is used as the second comprehensive score for each news item. After obtaining the first and second comprehensive scores for each news item, these two scores are determined as the news score for that news item. In other words, the same news item has two scoring systems. Subsequently, when recommending news based on these scores, at least two news items can be recommended to the user. The recommended news not only closely relates to the user's habits but also meets the user's new needs.

[0079] Optionally, based on the above method embodiments, the above-mentioned method for obtaining the attribute information of keywords in the keyword list and determining the keyword score of the keywords based on the attribute information includes the following schemes:

[0080] The keywords in the above keyword list are divided into a first part of keywords and a second part of keywords.

[0081] Obtain the attribute information of the keywords in the first part above;

[0082] Filter the minimum value in the above browsing data to obtain intermediate data;

[0083] Sort the values ​​in the above intermediate data to obtain sorted data, and calculate the set quantiles in the above sorted data;

[0084] Adjust the values ​​in the sorted data that are greater than or equal to the set quantile to the set quantile to obtain the preprocessed data;

[0085] The preprocessed data is standardized to obtain the target data;

[0086] Based on the above time data and the above target data, determine the first score of the keywords in the first part above;

[0087] Based on the model-based collaborative filtering recommendation algorithm, the aforementioned time data, the aforementioned user information, the aforementioned first part of keywords, and the aforementioned first score, the second score of the aforementioned second part of keywords is determined.

[0088] Based on the first score and the second score mentioned above, the scores for the keywords are determined.

[0089] First, score a small number of keywords in the first part. Then, based on the model, score the majority of keywords in the second part using the keywords in the first part. This can improve the scoring efficiency of keywords.

[0090] In an exemplary embodiment, after constructing a keyword list for each news item to be recommended, all keywords in the keyword list are divided into a first part of keywords and a second part of keywords. The number of keywords in the first part is less than the number of keywords in the second part. The attribute information of the keywords in the first part is the attribute information of the keywords for each news item to be recommended within a certain time period, and the attribute information of the keywords in the second part is the attribute information of the keywords for each news item to be recommended within other time periods. For example, the attribute information of the keywords in the first part is the attribute information of the keywords for news item 1 to be recommended between 10:00 and 12:00, and the attribute information of the keywords in the second part is the attribute information of the keywords for news item 1 to be recommended between 1:00 and 8:00 and between 1:00 and 24:00. The attribute information includes time data generated from browsing the news, user information, and browsing data. Browsing data includes the number of views in the last set number of days, the average daily number of views in the last set number of days, the average page dwell time in the last set number of days, the number of views on a statistical day, and the average page dwell time on a statistical day. The number of views in the last set number of days and the average daily number of views in the last set number of days are numerical values ​​from the browsing data. Time data includes the browsing date and the time period during which the news was viewed. The time period is represented by an hour ID, and user information is represented by a user ID.

[0091] Considering that the actual browsing data often has some problems, such as extreme values ​​(extremely large or extremely small values), and that some discrete values ​​need to be transformed before they can be used, numerical data also needs to be standardized to filter numerical values ​​and remove abnormal values.

[0092] After obtaining the browsing data for the first part of the keywords, the minimum value of the same type in the browsing data for each keyword in the first part of the keywords is removed. This yields the intermediate data for each keyword in the first part of the keywords, where the minimum value of the same type is not included in the intermediate data. For example, removing the minimum value from the browsing volume of all keywords in the first part of the keywords within a set number of days will result in intermediate data that does not include the minimum value of browsing volume within the set number of days.

[0093] After obtaining the intermediate data corresponding to each keyword, sort the values ​​of the same type in the intermediate data corresponding to all keywords in ascending order to obtain sorted data, that is, the values ​​in the sorted data are arranged in ascending order.

[0094] After obtaining the sorted data, a set quantile is calculated, for example, the 95th quantile. Next, all values ​​in the sorted data greater than or equal to this set quantile are adjusted to this quantile; that is, the set quantile is assigned to the values ​​in the sorted data that are greater than or equal to this set quantile, thus obtaining the preprocessed data. For example, if the set quantile is the 95th quantile, the formula for generating the preprocessed data is as follows, where X represents the value in the sorted data, Y1 represents the value in the preprocessed data, and X... 0.95 Representing the 95th percentile:

[0095]

[0096] Suppose that the 95th percentile of the sorted data is 95.5, and the values ​​in the sorted data that are greater than 95.5 are 96, 97, and 98, and the remaining values ​​in the sorted data are all less than 95.5. Then, we adjust 96, 97, and 98 to 95.5 to obtain the preprocessed data. In this case, the values ​​in the preprocessed data are all less than or equal to 95.5.

[0097] After obtaining the preprocessed data, a standardization function is used to standardize the values ​​in the preprocessed data to obtain the target data for the first part of the keywords. The standardization function is as follows:

[0098]

[0099] Here, Y2 represents the numerical value in the target data, and u is a fixed constant, the value of which is the mean of X. Standardizing the target data can classify its range into the interval [0,1] and provide optimal discrimination for X.

[0100] After obtaining the target data for the first part of the keywords, a comprehensive evaluation model, time data, and target data are used to score the first part of the keywords, resulting in the first score for the first part of the keywords.

[0101] For the comprehensive evaluation model, each day is divided into 24 hours, and personalized recommendations are made for each hourly segment. For news, before using the ALS collaborative filtering algorithm, it's necessary to obtain the user's rating for the news keywords they've visited. Based on this rating, the ALS collaborative filtering algorithm can be used to select the attribute information of multiple news keywords. These attributes reflect, to some extent, the user's preference for the corresponding news keywords. To synthesize multiple attribute information to give each news keyword a comprehensive score, a comprehensive evaluation model is used. Each attribute is assigned a corresponding weight, and then the weights are summed to obtain the total score. Unlike traditional comprehensive evaluation models, this model uses the attribute weights as hyperparameters for subsequent algorithms, obtaining the optimal weights through the model's training and parameter tuning process. This method is used because the attribute weights in traditional comprehensive evaluation models are obtained based on the relative importance values ​​between attributes. However, in the application scenario disclosed here, the importance of each attribute is not intuitively clear, so the weights are determined based on the model's parameter tuning and optimization process. On the other hand, data from the user within the last set number of days (e.g., 20 days) is selected from the time dimension. Taking into account the length of time since the recommendation date, a comprehensive score is given to the user based on their values ​​in various attribute information. In the time dimension, the scoring is performed according to the NDCG (Normalized Discounted Cumulative Gain) model, with attributes closer to the recommendation date having higher weights and vice versa. The comprehensive evaluation model is a function of the comprehensive evaluation, and its expression is as follows:

[0102]

[0103] Where Score represents the keyword score, m represents the number of days, n represents the date number, and X represents the keyword score. n,j w represents the value of the j-th attribute on day n. j This represents the weight of the j-th attribute; m and n are obtained from the time data of the first part of the keywords, X n,j The data is obtained from the target data of the first part of the keywords.

[0104] Given the large number of keywords and users in the news articles, model-based collaborative filtering is the most efficient approach. Specifically, the ALS parallel recommendation algorithm provided by the SAPK platform can be used to calculate the second score for the second set of keywords. The ALS parallel recommendation algorithm is a model-based recommendation algorithm, specifically based on matrix factorization.

[0105] After obtaining the first score for the first set of keywords, the model-based collaborative filtering recommendation algorithm, time data, user information, the first set of keywords, and the first score are used to calculate the second score for the second set of keywords. It should be understood that the hour ID can be obtained from the time data, and the user ID can be obtained from the user information. The user ID, hour ID, the first set of keywords, and the first score for the first set of keywords are input into the model-based collaborative filtering recommendation algorithm. This algorithm outputs the time IDs for other time periods and the second score for the second set of keywords. Specifically, the input format of this algorithm is (user ID_hour ID, first set of keywords, first score), and the output formula is (time IDs for other time periods, second score for the second set of keywords), meaning that the second score for the second set of keywords can be obtained through the output of this algorithm.

[0106] After obtaining the first score of the first part of the keywords and the second score of the second part of the keywords, the first score of the first part of the keywords and the second score of the second part of the keywords constitute the keyword score of all keywords in the keyword list. In this way, the keyword score of all keywords in each news article to be recommended can be obtained efficiently.

[0107] Optionally, based on the above method embodiments, S133 includes the following solutions:

[0108] If the above keyword list contains the above frequent keyword combinations, the keyword scores of the above keywords included in the above frequent keyword combinations are summed to obtain the keyword combination score of the above frequent keyword combinations.

[0109] When recommending news to users with ample information, a scoring system is used to evaluate each frequent keyword combination based on the keywords in the keyword list. Specifically, the system iterates through the preferred keywords in each frequent keyword combination, checking if all preferred keywords in that combination are also present in the keyword list. If they are, the keyword list contains the frequent keyword combination. For example, if keyword list 1 contains A, B, C, D, and E, and frequent keyword combination 1 contains A, B, and C, then keyword list 1 contains frequent keyword combination 1.

[0110] When the keyword list contains all preferred keywords in frequent keyword combinations, the keyword score corresponding to the target keyword is obtained from the keyword list. The target keyword is the same keyword in the keyword list that is included in the frequent keyword combination. Then, the keyword scores corresponding to the target keyword are summed to obtain a total value, which is then used as the keyword combination score for the frequent keyword combination. For example, if frequent keyword combination 1 includes keywords A and B, and A and B are in the keyword list of the news item to be recommended, and both A and B have a keyword score of 1, then the keyword combination score for frequent keyword combination 1 is 2 points. Since the keyword score is calculated based on the keyword attribute information, and attribute information can increase the amount of data used to score frequent keyword combinations, using keyword scores to score frequent keyword combinations can improve the accuracy of keyword combination scoring.

[0111] Optionally, based on the above method embodiments, S140 includes the following solutions:

[0112] Based on the order of the first comprehensive score in the above news scoring, the first candidate news is determined from all the above news to be recommended.

[0113] Based on the ranking of the second comprehensive score in the above news scoring, the second alternative news was determined from all the above news to be recommended.

[0114] The aforementioned first and second alternative news items are selected as the target news items to be recommended to the aforementioned target users.

[0115] When recommending news to users with ample information, the process is as follows: First, based on the first comprehensive score of all the news items to be recommended, ranked from highest to lowest, the top N news items with the highest first comprehensive score are selected. These N selected news items are the first candidate news items to recommend to the target user. Here, N is greater than or equal to 1. For example, N=3 means selecting the top 3 news items with the highest first comprehensive score as the first candidate news items to recommend to the target user. Second, based on the second comprehensive score of all the news items to be recommended, ranked from highest to lowest, the top M news items with the highest second comprehensive score are selected. These M selected news items are the second candidate news items to recommend to the target user. Here, M is greater than or equal to 1. For example, M=2 means selecting the top 2 news items with the highest second comprehensive score as the second candidate news items to recommend to the target user. After obtaining the first and second candidate news items, these two news items are used as the target news items to recommend to the target user; that is, the target news items must include at least 2 news items from the first candidate news items.

[0116] After scoring each news item to be recommended, for the same target user, each news item will have two scores: a first comprehensive score and a second comprehensive score. Since the first comprehensive score is obtained based on the attribute information of the keywords contained in the news item, and these keywords are obtained using the ALS collaborative filtering algorithm, the first comprehensive score can be understood as a news score based on the ALS collaborative filtering algorithm. Furthermore, since the frequent keyword combinations are obtained using the FPGrowth association rule algorithm, and the second comprehensive score is obtained by scoring the keyword combinations of frequent keyword combinations, the second comprehensive score can be understood as a news score based on the FPGrowth association rule algorithm. Because the FPGrowth association rule algorithm and the ALS collaborative filtering algorithm have different focuses—FPgrowth primarily mines users' existing habits (e.g., users' habitual search keywords), while ALS collaborative filtering helps to uncover new user needs and can recommend novel news—the two algorithms can be combined and complement each other. By recommending news based on both algorithms, the accuracy of news recommendations can be improved, the recommendation effect enhanced, and user needs better met.

[0117] For example, such as Figure 4 As shown, Figure 4 This illustration shows another exemplary flowchart of S130 in the news recommendation method provided in this embodiment. Optionally, based on the above method embodiment, S130 includes the following:

[0118] S1321: Check whether the keywords in the above keyword list have corresponding attribute information;

[0119] S1322: If the keywords in the keyword list do not have corresponding attribute information, and the keyword list contains the frequent keyword combination, determine the keyword combination score of the frequent keyword combination based on the number of keywords contained in the frequent keyword combination; or, set a preset score for the frequent keyword combination to obtain the keyword combination score.

[0120] S1323: Determine the news score for each of the above-mentioned news items to be recommended based on the sum of the scores of the above-mentioned keyword combinations.

[0121] In one exemplary embodiment, a scoring method for calculating the keyword combination score can be obtained by detecting whether the keywords in the keyword list correspond to attribute information. If the keywords in the keyword list do not correspond to attribute information, it means that sufficient data that can participate in the scoring of the keyword combination cannot be obtained from the relevant information corresponding to the news previously viewed by the user. However, a small amount of data that can participate in the scoring of the keyword combination can be obtained. The data that can be obtained is the keywords in each of the news articles to be recommended and the keyword combination corresponding to the target user. This is a case of recommending news to users with limited information.

[0122] Furthermore, when the keyword list contains frequent keyword combinations, the number of keywords contained in each frequent keyword combination is obtained. This number is also the total number of preferred keywords in the frequent keyword combination, and all preferred keywords in the frequent keyword combination are included in the keyword list. The total number of preferred keywords is then determined as the keyword combination score for that frequent keyword combination. For example, if the total number of preferred keywords is 5, the keyword combination score for that frequent keyword combination is 5 points. Alternatively, when the keyword list contains frequent keyword combinations, a preset score is set for each frequent keyword combination to obtain its keyword combination score. For example, if the preset score for each frequent keyword combination is 1 point, then the keyword combination score for each frequent keyword combination is 1 point.

[0123] Optionally, based on the above method embodiments, S140 includes the following solutions:

[0124] Determine the ranking order of all the news ratings for the above-mentioned news items to be recommended;

[0125] Based on the above sorting order, target news to be recommended to the target users is determined from all the news to be recommended.

[0126] Based on the scenario of recommending news to users with limited information, after obtaining the news rating for each news item to be recommended, the top M news items with the highest news ratings are selected from all the news items to be recommended, arranged in descending order of their ratings. These M selected news items are the target news items to be recommended to the target user. M can be greater than or equal to 1, or they can be different. For example, M = 2 means that the top two news items with the highest news ratings are selected as the target news items to be recommended to the target user.

[0127] Since the frequent keyword combinations are obtained through the FPGrowth association rule algorithm, which primarily mines users' existing habits (e.g., frequently used search keywords), it leverages established user habits. By scoring each news item to be recommended using the frequent keyword combinations mined by the FPGrowth algorithm and then recommending the news, the accuracy of news recommendations can be improved. This effectively avoids the problem of poor recommendation performance when using traditional ALS recommendation algorithms because of insufficient relevant historical information from users.

[0128] Optionally, based on the above method embodiments, after S140, the above news recommendation method further includes the following:

[0129] Obtain the preferred times when the target users previously browsed news;

[0130] Divide each day into multiple time periods, determine the time periods where the above-mentioned preferred times fall, and obtain the recommended time periods;

[0131] Based on the aforementioned recommended time period, the aforementioned target news will be recommended to the aforementioned target users.

[0132] Based on whether news recommendations are made to users with ample or limited information, after identifying the target news to recommend to the target user, the system obtains the target user's past preferred news browsing times. These preferred times reflect the specific time periods during which the target user typically enjoys or habitually browses news. Then, the day is divided into multiple time periods, for example, dividing a 24-hour day into 24 time periods. The system then determines which specific time period the preferred time falls within, thus obtaining the recommended time period. For instance, if the preferred time is 10:30 AM, and 10:30 AM falls within the 10:00-11:00 AM time period, then 10:00-11:00 AM is the recommended time period. Therefore, the target news will be recommended to the target user between 10:00 AM and 11:00 AM every subsequent day, thus achieving personalized news recommendations based on hourly time periods.

[0133] Based on the above embodiments, this disclosure provides news recommendations to users based on keywords. To more accurately utilize keywords for news recommendations, it is necessary to mine the association rules between keywords and use frequently occurring keyword combinations for news recommendations to achieve higher accuracy. The recommendation methods vary depending on the user:

[0134] For new users, based on their biometric information, we can identify the trending news topics that their group belongs to and then recommend trending news to them, which can effectively solve the cold start problem.

[0135] For users with limited information, using frequent keyword combinations mined by the FPgrowth association rule algorithm for news recommendations can improve the accuracy of news recommendations. This can effectively avoid the problem of poor recommendation results when using traditional ALS recommendation algorithms for news recommendations when users lack sufficient effective historical information.

[0136] For users with ample information, on the one hand, the FPGrowth association rule algorithm mines existing user habits (e.g., frequently used search keywords) to uncover established user habits; on the other hand, the ALS recommendation algorithm uncovers new user needs, allowing for the recommendation of novel news. Therefore, combining these two algorithms, and recommending news based on each, improves the accuracy and effectiveness of news recommendations, ultimately better meeting user needs.

[0137] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0138] Figure 5 A schematic diagram of a news recommendation device that can be applied to an embodiment of this disclosure is shown. Please see below. Figure 5 The news recommendation device shown in the figure can be implemented as a whole or part of an electronic device through software, hardware, or a combination of both, or it can be integrated as an independent module into an electronic device or server.

[0139] The news recommendation device 500 in this embodiment includes:

[0140] List building module 510 is used to build a list of keywords for each news item to be recommended;

[0141] The combination acquisition module 520 is used to acquire frequent keyword combinations generated from the target user's past browsing of news;

[0142] The news scoring module 530 is used to score each of the above-mentioned news items to be recommended based on the keywords in the keyword list and the combination of the most frequent keywords, so as to obtain the news score of each of the above-mentioned news items to be recommended.

[0143] The news selection module 540 is used to determine the target news to recommend to the target user from all the above-mentioned news to be recommended based on the news rating of each of the above-mentioned news to be recommended.

[0144] In an exemplary embodiment, based on the foregoing scheme, the list construction module 510 includes:

[0145] The word segmentation unit is used to segment the text of each of the above-mentioned news articles to be recommended, resulting in multiple words;

[0146] The weight setting unit is used to set a weight value for each of the above word segments using the inverse document frequency algorithm;

[0147] The keyword determination unit is used to select the word segments with weight values ​​greater than preset values ​​as keywords for each of the above-mentioned news items to be recommended.

[0148] The list construction unit is used to build the keyword list based on the keywords of each of the above-mentioned news items to be recommended.

[0149] In an exemplary embodiment, based on the foregoing scheme, the news rating module 530 includes:

[0150] The first detection unit is used to detect whether the keywords in the keyword list correspond to attribute information.

[0151] The keyword scoring unit is used to obtain the attribute information of the keyword in the keyword list when the keyword in the keyword list has corresponding attribute information, and to determine the keyword score of the keyword based on the attribute information.

[0152] The first combination scoring unit is used to determine the keyword combination score of the above-mentioned keyword frequent item combination based on the above-mentioned keyword scores of the above-mentioned keywords in the above-mentioned keyword list.

[0153] The news scoring calculation unit is used to determine a first comprehensive score for each of the above-mentioned news items to be recommended based on the sum of the scores of the above-mentioned keywords, and to determine a second comprehensive score for each of the above-mentioned news items to be recommended based on the sum of the scores of the above-mentioned keyword combinations.

[0154] The first news scoring unit is used to determine the news score for each of the aforementioned news items to be recommended based on the aforementioned first comprehensive score and the aforementioned second comprehensive score.

[0155] In an exemplary embodiment, based on the foregoing scheme, the keyword scoring unit, in obtaining the attribute information of the keywords in the keyword list and determining the keyword score of the keywords based on the attribute information, includes:

[0156] A segmentation unit is used to divide the keywords in the keyword list into a first part of keywords and a second part of keywords.

[0157] The information acquisition subunit is used to acquire the attribute information of the keywords in the first part mentioned above; wherein, the attribute information includes time data generated from browsing news, user information, and browsing data;

[0158] The filtering subunit is used to filter the minimum value in the above browsing data to obtain intermediate data;

[0159] The sorting subunit is used to sort the values ​​in the intermediate data to obtain sorted data and calculate the set quantiles in the sorted data.

[0160] The adjustment sub-unit is used to adjust the values ​​in the sorted data that are greater than or equal to the set quantile to the set quantile, so as to obtain preprocessed data;

[0161] The transformation subunit is used to perform standard transformation on the preprocessed data to obtain the target data.

[0162] The first scoring subunit is used to determine the first score of the keywords in the first part based on the time data and the target data mentioned above.

[0163] The second scoring subunit is used to determine the second score of the second part of the keywords based on the model-based collaborative filtering recommendation algorithm, the aforementioned time data, the aforementioned user information, the aforementioned first part of keywords, and the aforementioned first score.

[0164] The scoring determination sub-unit is used to determine the keyword scores based on the first score and the second score mentioned above.

[0165] In an exemplary embodiment, based on the foregoing scheme, the first combination scoring unit is specifically used to accumulate the keyword scores of the keywords included in the frequent keyword combination when the keyword list contains the frequent keyword combination, to obtain the keyword combination score of the frequent keyword combination.

[0166] In an exemplary embodiment, based on the foregoing solution, the news selection module 540 includes:

[0167] The first news selection unit is used to determine the first candidate news from all the news to be recommended based on the order of the first comprehensive score in the news scoring.

[0168] The second news selection unit is used to determine the second candidate news from all the news to be recommended based on the order of the second comprehensive score in the news scoring.

[0169] The target news determination unit is used to determine the first alternative news and the second alternative news as the target news to be recommended to the target user.

[0170] In an exemplary embodiment, based on the foregoing solution, the news rating module 530 further includes:

[0171] The second detection unit is used to detect whether the keywords in the keyword list correspond to attribute information.

[0172] The second combination scoring unit is used to determine the keyword combination score of the frequent keyword combination based on the number of keywords contained in the frequent keyword combination when the keywords in the keyword list do not have corresponding attribute information and the keyword list contains the frequent keyword combination; or, to set a preset score for the frequent keyword combination to obtain the keyword combination score.

[0173] The second news scoring unit is used to determine the news score for each of the above-mentioned news items to be recommended based on the sum of the scores of the above-mentioned keyword combinations.

[0174] In an exemplary embodiment, based on the foregoing scheme, the combination acquisition module 520 is specifically used to acquire the above-mentioned frequent item combination of keywords generated by the target user's previous browsing of news using a frequent itemset mining algorithm.

[0175] In an exemplary embodiment, based on the foregoing solution, the news recommendation device further includes:

[0176] The time acquisition unit is used to acquire the preferred times when the target user previously browsed news.

[0177] The time period determination unit is used to divide the day into multiple time periods, determine the time period in which the above-mentioned preferred time period falls, and obtain the recommended time period.

[0178] The news recommendation unit is used to recommend the aforementioned target news to the aforementioned target users based on the aforementioned recommendation time period.

[0179] It should be noted that the news recommendation device provided in the above embodiments is only illustrated by the division of each functional module. In practical applications, the functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the news recommendation device and the news recommendation method embodiments provided in the above embodiments belong to the same concept. Therefore, for details not disclosed in the device embodiments of this disclosure, please refer to the above-described embodiments of the news recommendation method, which will not be repeated here.

[0180] The sequence numbers of the embodiments disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0181] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described in the foregoing embodiments. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and data.

[0182] This disclosure also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the methods described above.

[0183] Figure 6 A schematic diagram of the electronic device is shown. Please refer to [link / reference]. Figure 6 As shown, the electronic device 600 includes a processor 601 and a memory 602.

[0184] In this embodiment, processor 601 is the control center of the computer system and can be a processor of a physical machine or a processor of a virtual machine. Processor 601 may include one or more processing cores, such as a 4-core processor or an 8-core processor. Processor 601 can be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 601 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0185] In this embodiment of the disclosure, the processor 601 is specifically configured to: construct a keyword list for each news item to be recommended; obtain attribute information of the keywords in the keyword list, and determine the keyword score of the keywords based on the attribute information; wherein the attribute information is generated after news containing the keywords is viewed; obtain frequent keyword combinations generated by the target user's previous browsing of news; determine the keyword combination score of the frequent keyword combinations based on the inclusion relationship between the keyword list and the frequent keyword combinations; and determine the target news to be recommended to the target user from all the news items to be recommended based on the keyword score and the keyword combination score.

[0186] Furthermore, the processor 601 is also configured to: construct a keyword list for each news item to be recommended; obtain frequent keyword combinations generated by the target user's previous browsing of news; score each news item to be recommended based on the keywords in the keyword list and the frequent keyword combinations, thereby obtaining a news score for each news item to be recommended; and determine the target news item to be recommended to the target user from all the news items to be recommended based on the news score for each news item to be recommended.

[0187] Furthermore, the processor 601 is also used to: perform word segmentation on the text of each of the above-mentioned news articles to be recommended, to obtain multiple word segments; set a weight value for each of the above-mentioned word segments using a term frequency inverse document frequency algorithm; use the above-mentioned word segments with a weight value greater than a preset value as keywords for each of the above-mentioned news articles to be recommended; and construct the above-mentioned keyword list based on the keywords of each of the above-mentioned news articles to be recommended.

[0188] Furthermore, the processor 601 is also configured to: detect whether the keywords in the keyword list correspond to attribute information; if the keywords in the keyword list correspond to attribute information, obtain the attribute information of the keywords in the keyword list, and determine the keyword score of the keywords based on the attribute information; determine the keyword combination score of the frequent keyword combination based on the keyword scores of the keywords in the keyword list; determine a first comprehensive score for each of the news items to be recommended based on the sum of the keyword scores, and determine a second comprehensive score for each of the news items to be recommended based on the sum of the keyword combination scores; and determine a news score for each of the news items to be recommended based on the first comprehensive score and the second comprehensive score.

[0189] Furthermore, the processor 601 is also configured to: divide the keywords in the keyword list into a first part of keywords and a second part of keywords; obtain attribute information of the first part of keywords; wherein the attribute information includes time data generated from browsing news, user information, and browsing data; filter the minimum value in the browsing data to obtain intermediate data; sort the values ​​in the intermediate data to obtain sorted data, and calculate a set quantile in the sorted data; adjust the values ​​in the sorted data that are greater than or equal to the set quantile to the set quantile to obtain preprocessed data; standardize the preprocessed data to obtain target data; determine a first score for the first part of keywords based on the time data and the target data; determine a second score for the second part of keywords based on a model-based collaborative filtering recommendation algorithm, the time data, the user information, the first part of keywords, and the first score; and determine the keyword score based on the first score and the second score.

[0190] Furthermore, the processor 601 is also configured to: when the keyword list contains the frequent keyword combination, sum the keyword scores of the keywords included in the frequent keyword combination to obtain the keyword combination score of the frequent keyword combination.

[0191] Furthermore, the processor 601 is also configured to: determine a first candidate news item from all the news items to be recommended based on the order of the first comprehensive score in the news rating; determine a second candidate news item from all the news items to be recommended based on the order of the second comprehensive score in the news rating; and determine the first candidate news item and the second candidate news item as the target news item to be recommended to the target user.

[0192] Furthermore, the processor 601 is also configured to: detect whether the keywords in the keyword list correspond to attribute information; if the keywords in the keyword list do not correspond to attribute information, and the keyword list contains frequent keyword combinations, determine the keyword combination score of the frequent keyword combination based on the number of keywords contained in the frequent keyword combination; or, set a preset score for the frequent keyword combination to obtain the keyword combination score; and determine the news score for each of the recommended news items based on the sum of the keyword combination scores.

[0193] Furthermore, the processor 601 is also used to: employ a frequent itemset mining algorithm to obtain the frequent item combinations of the keywords generated by the target user's previous browsing of news.

[0194] Furthermore, the processor 601 is also used to: obtain the preferred time of the target user's past browsing of news; divide the time of each day into multiple time periods, determine the time period in which the preferred time is located, and obtain the recommended time period; and recommend the target news to the target user based on the recommended time period.

[0195] Memory 602 may include one or more computer-readable storage media, which may be non-transitory. Memory 602 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments of this disclosure, the non-transitory computer-readable storage media in memory 602 are used to store at least one instruction for execution by processor 601 to implement the methods in the embodiments of this disclosure.

[0196] In some embodiments, the electronic device 600 further includes a peripheral device interface 603 and at least one peripheral device. The processor 601, memory 602, and peripheral device interface 603 are connected via a bus or signal line. Each peripheral device is connected to the peripheral device interface 603 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of a display screen 604, a camera 605, and an audio circuit 606.

[0197] Peripheral device interface 603 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 601 and memory 602. In some embodiments of this disclosure, processor 601, memory 602, and peripheral device interface 603 are integrated on the same chip or circuit board; in other embodiments of this disclosure, any one or two of processor 601, memory 602, and peripheral device interface 603 can be implemented on separate chips or circuit boards. This disclosure does not specifically limit the scope of the embodiments.

[0198] Display screen 604 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 604 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 601 for processing. In this case, display screen 604 can also be used to provide virtual buttons and a virtual keyboard, also known as soft buttons and soft keyboards. In some embodiments of this disclosure, there may be one display screen 604, which serves as the front panel of electronic device 600; in other embodiments, there may be at least two display screens 604, respectively disposed on different surfaces of electronic device 600 or in a folded design; in still other embodiments, display screen 604 may be a flexible display screen, disposed on a curved or folded surface of electronic device 600. Furthermore, display screen 604 may be configured as a non-rectangular irregular shape, i.e., a non-rectangular screen. Display screen 604 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0199] Camera 605 is used to capture images or videos. Optionally, camera 605 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the electronic device 600, and the rear-facing camera is located on the back of the electronic device 600. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments of this disclosure, camera 605 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cool light flash, which can be used for light compensation at different color temperatures.

[0200] The audio circuit 606 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to the processor 601 for processing. For stereo sound acquisition or noise reduction purposes, there may be multiple microphones, each located in a different part of the electronic device 600. The microphone may also be an array microphone or an omnidirectional microphone.

[0201] Power supply 607 is used to power each component in electronic device 600. Power supply 607 can be alternating current, direct current, a disposable battery, or a rechargeable battery. When power supply 607 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, and a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0202] The structural block diagram of the electronic device 600 shown in this embodiment does not constitute a limitation on the electronic device 600. The electronic device 600 may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0203] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this specification are all authorized by the user or fully authorized by each party, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the object characteristics, interactive behavior characteristics, and user information involved in this specification were all obtained with full authorization.

[0204] In the description of this disclosure, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this disclosure based on the specific circumstances. Furthermore, in the description of this disclosure, unless otherwise stated, "multiple" means two or more. "And" describes the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0205] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, equivalent variations made in accordance with the claims of this disclosure are still within the scope of this disclosure.

Claims

1. A news recommendation method, characterized in that, The news recommendation method includes: Build a keyword list for each news item to be recommended; Obtain frequent keyword combinations generated from the target user's past news browsing history; Based on the keywords in the keyword list and the frequent combinations of the keywords, each of the news items to be recommended is scored to obtain a news score for each of the news items to be recommended. Based on the news rating of each of the news items to be recommended, the target news items to be recommended to the target user are determined from all the news items to be recommended. The step of scoring each piece of news to be recommended based on keywords in the keyword list and combinations of frequent keyword terms, to obtain a news score for each piece of news to be recommended, includes: Detect whether the keywords in the keyword list correspond to attribute information; If the keywords in the keyword list correspond to attribute information, the keywords in the keyword list are divided into a first part of keywords and a second part of keywords. Obtain the attribute information of the first part of keywords; wherein, the attribute information includes time data generated from browsing news, user information, and browsing data; The minimum value in the browsing data is filtered to obtain intermediate data; The values ​​in the intermediate data are sorted to obtain sorted data, and a set quantile in the sorted data is calculated. The values ​​in the sorted data that are greater than or equal to the set quantile are adjusted to the set quantile to obtain preprocessed data; The preprocessed data is standardized to obtain the target data; Based on the time data and the target data, determine the first score of the first part of the keywords; A model-based collaborative filtering recommendation algorithm, the time data, the user information, the first part of keywords, and the first score are used to determine the second score of the second part of keywords; Based on the first score and the second score, determine the keyword score for the keyword; If the keyword list contains the frequent keyword combination, the keyword combination score of the frequent keyword combination is determined based on the keyword score of the keywords included in the frequent keyword combination; A first comprehensive score for each of the news items to be recommended is determined based on the sum of the keyword scores, and a second comprehensive score for each of the news items to be recommended is determined based on the sum of the keyword combination scores. Based on the first comprehensive score and the second comprehensive score, a news score is determined for each of the news items to be recommended.

2. The news recommendation method as described in claim 1, characterized in that, The steps for constructing the keyword list for each news item to be recommended include: Each of the news articles to be recommended is segmented into multiple words. A weight value is assigned to each word segmentation using a term frequency inverse document frequency algorithm. The word segments with weight values ​​greater than a preset value are used as keywords for each of the news articles to be recommended; The keyword list is constructed based on the keywords of each of the news items to be recommended.

3. The news recommendation method as described in claim 1, characterized in that, The step of determining the keyword combination score of the keyword frequent item combination based on the keyword scores of the keywords included in the keyword frequent item combination includes: The keyword scores of the keywords included in the frequent keyword combination are summed to obtain the keyword combination score.

4. The news recommendation method as described in claim 1, characterized in that, The step of determining the target news to recommend to the target user from all the news to be recommended based on the news rating of each news item to be recommended includes: The first candidate news is determined from all the news to be recommended based on the order of the first comprehensive score in the news rating. The second candidate news is determined from all the news to be recommended based on the ranking order of the second comprehensive score in the news scoring. The first and second candidate news items are selected as the target news items to be recommended to the target user.

5. The news recommendation method as described in claim 1, characterized in that, The step of scoring each piece of news to be recommended based on keywords in the keyword list and combinations of frequent keyword terms, to obtain a news score for each piece of news to be recommended, includes: Detect whether the keywords in the keyword list correspond to attribute information; If a keyword in the keyword list does not have corresponding attribute information, and the keyword list contains frequent keyword combinations, the keyword combination score of the frequent keyword combination is determined based on the number of keywords contained in the frequent keyword combination; or, a preset score is set for the frequent keyword combination to obtain the keyword combination score. The news score for each of the recommended news items is determined based on the sum of the scores for the keyword combinations.

6. The news recommendation method as described in claim 1, characterized in that, The steps for obtaining frequent keyword combinations generated from the target user's past news browsing include: A frequent itemset mining algorithm is used to obtain the frequent item combinations of keywords generated from the target user's past browsing of news.

7. The news recommendation method as described in any one of claims 1 to 6, characterized in that, After the step of determining the target news to recommend to the target user from all the news to be recommended based on the news rating of each news to be recommended, the news recommendation method further includes: Obtain the target user's past preferred times for browsing news; The day is divided into multiple time periods, and the time period in which the preferred time is located is determined to obtain the recommended time period. The target news is recommended to the target user based on the recommended time period.

8. A news recommendation device, characterized in that, The news recommendation device includes: The list building module is used to build a list of keywords for each news item to be recommended; The combination acquisition module is used to acquire frequent keyword combinations generated from the target user's past browsing of news; The news scoring module is used to detect whether the keywords in the keyword list correspond to attribute information; if the keywords in the keyword list correspond to attribute information, the keywords in the keyword list are divided into a first part of keywords and a second part of keywords; the attribute information of the first part of keywords is obtained; wherein, the attribute information includes time data, user information, and browsing data generated by browsing news; the minimum value in the browsing data is filtered to obtain intermediate data; the values ​​in the intermediate data are sorted to obtain sorted data, and a set quantile is calculated in the sorted data; the values ​​in the sorted data that are greater than or equal to the set quantile are adjusted to the set quantile to obtain preprocessed data; the preprocessed data is standardized to obtain target data; based on the time data and the... The target data is used to determine a first score for the first part of the keywords; based on the collaborative filtering recommendation algorithm of the model, the time data, the user information, the first part of the keywords, and the first score, a second score for the second part of the keywords is determined; based on the first score and the second score, a keyword score for the keyword is determined; if the keyword list contains the frequent keyword combination, a keyword combination score for the frequent keyword combination is determined based on the keyword scores of the keywords included in the frequent keyword combination; a first comprehensive score for each news item to be recommended is determined based on the sum of the keyword scores, and a second comprehensive score for each news item to be recommended is determined based on the sum of the keyword combination scores; a news score for each news item to be recommended is determined based on the first comprehensive score and the second comprehensive score. The news selection module is used to determine the target news to recommend to the target user from all the news to be recommended based on the news rating of each news item to be recommended.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the news recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the news recommendation method as described in any one of claims 1 to 7.

Images