TW201706880A - News following and recommendation method providing a personalized news following and recommendation method in a social network environment - Google Patents

Abstract

A news following and recommendation method comprises the following steps implemented from a user end: (a) obtaining several corresponding first news articles from each news server; (b) based on several second news articles, labeling each first news article as one of several historical events or a new event; (c) obtaining a user data from a community server; (d) based on the user data, obtaining a browsing similarity between the first user and each second user; (e) based on the browsing similarity, selecting one second user with the browsing similarity greater than a browsing threshold value; (f) based on the user data, obtaining several ratings on the historical events and the new event from the first user and the selected second user; and (g) based on the rating and the browsing similarity, obtaining at least one recommended event.

Description

News tracking and recommendation methods

The invention relates to a news recommendation method, in particular to a personalized news tracking and recommendation method in a social network environment.

With the development of information and the Internet, most people in recent years have been concerned about news events through online news, or shared and published their views on news events through social networking sites. Today, many journalists provide online news through a number of news websites they have established, such as Apple Daily, People's Daily News, Central News, News, and China News Network. Users often spend a lot of time searching for news events they care about. It is necessary to seek a news tracking and recommendation method to reduce the time it takes for users to search and track news.

Most of the existing news recommendation methods are to search for the focus news of the day according to a plurality of keywords by a processing unit, and provide the searched focus news to the user as at least one recommended news. Another recommended method is to obtain a popular news article by the processing unit according to the number of positive comments in the online news article, and provide the obtained popular news article to the user as at least one recommended news. Of course However, none of the above news recommendation methods considers the user's tendency to press preferences and cannot provide personalized news recommendations for the user.

Accordingly, it is an object of the present invention to provide a news tracking and recommendation method that provides personalized recommendations.

Therefore, the news tracking and recommendation method is implemented by a user connected to a plurality of news servers and a community server via a network, and the news tracking and recommendation method comprises the following steps: (A) Each of the news servers corresponding to a news website obtains a plurality of corresponding first news articles related to a first time interval; (B) corresponding to a second time prior to the first time interval a second news article belonging to a plurality of different historical events, each first news article being marked as one of the historical events and a new event; (C) corresponding to a first user U ¹ and a plurality of the first user associated with the second user U ¹ U ² of the community server end, obtaining an associated user U ¹ to the first and second user U such user ² information; (D) according to the user data, the first user U ¹ is obtained with respect to each of the second user a browsing similarity S _i for the historical events and the new event N ; (E) selecting each second user according to each browsing similarity S _i a second user whose browsing similarity S _{i is} greater than a browsing threshold (F) obtaining each second user based on the user profile An evaluation value of N _k relative to each of the historical events and the new event U ¹ and the first phase a user evaluation value for each of these historical events and the new event of the N _k And (G) the evaluation values E ¹ , E ² obtained according to step (F) and the first user U ¹ relative to each second user The browsing similarity S _j captures at least one of the historical events and the new event N as at least one recommended event corresponding to the first user U ¹ .

The effect of the present invention is that the corresponding first news articles are obtained from each news server by the use end, and each first news article is marked as one of the historical events and the new event, And obtaining, by the user terminal, the user data from the community server, and obtaining the browsing similarity S and the evaluation values E ¹ and E ² according to the user data, and further, according to the evaluation value E ¹ , E ² and the browsing similarity S take the recommendation event to recommend to the first user U ¹ . Among them, the recommendation event contains a plurality of news articles related to each other, which is convenient for the user to track the news.

1‧‧‧Use side

2‧‧‧ News Server

3‧‧‧Community server

4‧‧‧Network

51~59‧‧‧Steps

521~526‧‧‧ substeps

541~548‧‧‧substeps

Other features and effects of the present invention will be apparent from the embodiments of the drawings, in which: 1 is a block diagram showing one of the methods for implementing the news tracking and recommendation method of the present invention. The user terminal is connected to a plurality of news servers and a community server via a network; FIG. 2 is a flowchart illustrating the news tracking of the present invention. And an embodiment of the recommended method; FIG. 3 is a flowchart illustrating how the first end of the first news article is marked as one of a plurality of historical events and a new event; and FIG. 4 is a flowchart illustrating How the user obtains a browsing similarity of the first user relative to each of the second users for the historical events and the new event.

Referring to FIG. 1 and FIG. 2, an embodiment of the news tracking and recommendation method of the present invention is implemented by a user terminal 1 connected to a plurality of news server terminals 2 and a community server terminal 3 via a network 4. In this embodiment, the news tracking and recommendation method is used to recommend a news article whose language category is Chinese.

The news server 2 provides a plurality of different news websites, such as Apple Daily, People's Daily News, Central News, News, Dongsen News, and China News Network, and stores a number of news articles. Each news website contains a number of news pages related to such news articles.

The community server 3, such as Facebook or Twitter, corresponds to a first user U ¹ and a plurality of second users U ² related to the first user U ¹ , and User data related to the first user U ¹ and the second user U ² are stored.

The user terminal 1, such as a personal computer, a notebook computer, a tablet computer, or a smart phone, is installed with an application program including a plurality of software applications to implement the news tracking and recommendation method of the present invention.

Referring to Figures 1 and 2, an embodiment of the news tracking and recommending method of the present invention comprises the following steps.

In step 51, the user terminal 1 obtains a corresponding plurality of first news related to a first time interval from each news server 2 corresponding to a news website by using a Web Crawler technology. article.

It is worth mentioning that because the search engine optimization (SEO) rules of each news website are different, especially for each news website, a news title tag (Tag), news content tag The labels of the other items are different. As shown in Table 1, the news headline label of the Apple Daily is the h1 label, and the news headline label of the popular news is the span label. Therefore, for the SEO rules of different websites, the use side 1 utilizes Different web crawler rules get these first news articles from each news site.

In step 52, the user terminal 1 marks each first news article according to a plurality of second news articles corresponding to a second time interval preceding the first time interval and belonging to a plurality of different historical events respectively. For one of these historical events and a new event.

It is worth mentioning that the news articles stored in each news server 2 contain corresponding first news articles. The user terminal 1 also obtains, from the news server 2, the corresponding second news articles related to the second time interval. The news articles stored by each news server 2 also include corresponding second news articles. In this embodiment, the length of the second time interval is, for example, one week, and the length of the first time interval is, for example, one day. The user terminal 1 obtains the corresponding second news articles from each news server 2 in the second time interval.

It is worth mentioning that the user terminal 1 marks each of the first news articles as one of the historical events and the new event by performing the following sub-steps 521-526 (see FIG. 3).

In sub-step 521, the user terminal 1 divides each first news article into corresponding plurality of first words according to a word-breaking method, and divides each second news article into corresponding plurality of second words. word. In this embodiment, the method of word breaking is a method of word breaking related to Chinese Knowledge Information Processing (CKIP).

In sub-step 522, the user terminal 1 uses a word frequency reverse file frequency according to the first word corresponding to each first news article and the second word corresponding to each second news article (Term) The frequency-inverse document frequency (TF-IDF) method obtains a plurality of first keywords corresponding to the first news article and a plurality of second keywords corresponding to the second news article.

In sub-step 523, the user terminal 1 calculates the same first news article relative to each of the first keywords corresponding to each first news article and the second keywords corresponding to each second news article. A type of similarity of a second news article. In this embodiment, the user end 1 uses a cosine similarity formula to calculate the type similarity.

In sub-step 524, the user terminal 1 determines, according to the type of similarity corresponding to each first news article, whether one of the types of similarities corresponding to the first news article is greater than a type threshold. . When the determination is yes, sub-step 525 is performed, otherwise, sub-step 526 is performed.

In sub-step 525, the user terminal 1 marks the first news article as a historical event to which the second news article with the highest type similarity belongs according to the type of similarity corresponding to the first news article.

In sub-step 526, the consumer 1 marks the first news article as the new event.

In step 53, the user terminal 1 obtains a correlation from the community server 3 corresponding to a first user U ¹ and a plurality of second users U ² related to the first user U ¹ . User data of the first user U ¹ and the second user U ² .

In step 54, the usage information of the user terminal 1, the first user U ¹ is obtained with respect to each of the second user For the historical similarity and a browsing similarity S _{i of} the new event N , the number of the second users U ² is I, 1 i I.

It is worth mentioning that the end of a line by using the following sub-steps 541 to 548 (see FIG. 4), to obtain the first user U ¹ with respect to each of the second user A browsing similarity S _i for the historical events and the new event N.

In sub-step 541, the user terminal 1 obtains a plurality of user articles related to at least one of the first user U ¹ and the second users U ² according to the user profile.

In sub-step 542, the user terminal 1 divides each user article into corresponding plurality of user words according to a word-breaking method. In this embodiment, the word breaking method is also the word breaking method related to Chinese knowledge information processing.

In sub-step 543, the user terminal 1 obtains a plurality of user keywords corresponding to the user article by using the word frequency reverse file frequency method according to the user words corresponding to each user article.

In sub-step 544, the user terminal 1 is based on the user keywords corresponding to each user article, the first keywords corresponding to each first news article, and the corresponding corresponding to each second news article. The second keyword calculates an article similarity of the same user article relative to each of the first news articles and the second news articles.

In sub-step 545, the user terminal 1 determines whether one of the similarities of the articles corresponding to the user article is greater than a usage threshold according to the similarity of the articles corresponding to each user article. When the determination is yes, then sub-step 546 is performed, otherwise sub-step 547 is performed.

It is worth mentioning that the user articles contained in the user profile can be related to different topics such as food, travel or news. However, the user terminal 1 obtains the first user U ¹ relative to each second user. When the browsing similarity S _i is used, the user terminal 1 only needs to be obtained according to the user article related to the historical event or the new event N , and is not required to be independent of the historical event or the new event N (eg User articles related to food or travel notes are available. Therefore, the user 1 calculates the similarity of each article of each user article relative to each of the first news articles and the second news articles, and determines each of the calculated user articles. Whether one of the similarities of the articles is greater than the usage threshold, since the first news articles and the second news articles belong to articles related to the historical events or the new events N , When one of the similarities of the articles corresponding to each user article is greater than the usage threshold, the user 1 determines that the user article is related to the historical events and the new event N.

In sub-step 546, the user terminal 1 marks the user article as one of the historical events and the new event N according to the similarity of the articles corresponding to the user article, to obtain the first User U ¹ and the second user U ² are in a mapping table with respect to one of the historical events and the new event N. In this embodiment, the user end 1 marks the user article as one of the first and second news articles with the highest similarity of the article according to the similarity of the articles corresponding to the user article. One of the new events and historical events.

For example, if there are 7 such user articles A related to the news, the second user U ² has 6 persons, and the historical events and the new event N have 8 pieces, and the user articles are The relationship between A and the historical events and the new event N is: A ₁ is marked as N ₁ , A ₂ is marked as N ₂ , A ₃ is marked as N ₃ , A ₄ is marked as N ₄ , A ₅ is Marked as N ₅ , A ₆ is labeled N ₆ , A ₇ is labeled N ₇ , no user articles are labeled N ₈ , and the user articles A and the first user U ¹ and the like The relationship between the second user U ² is as shown in Table 2 below. When the first and second users U ¹ , U ² correspond to the value of the user article A, the first and second users U ¹ , U are represented. ² related to the user article A , when the first and second users U ¹ , U ² correspond to the user article A, the value of 0 represents the first and second users U ¹ , U ² and the user article A If not, the mapping table is as shown in Table 3 below. When the first and second users U ¹ and U ² correspond to the value of the event N, the first and second users U ¹ and U ^{2 are represented.} N associated with the event, when And a second user U ^1, U ² corresponds to the event N is 0, i.e. representing the first and second users U ^1, U ² and N independent events.

In sub-step 547, the consumer 1 determines that the user article is unrelated to the historical event or the new event N.

In sub-step 548, the terminal 1 is used to obtain the first user U ¹ phase using a formula based on the cosine similarity mapping relationship table, for each pair of the second user A browsing similarity S _i for the historical events and the new event N.

In step 55, the user terminal 1 selects each second user according to each browsing similarity S _i . a second user whose browsing similarity S _{i is} greater than a browsing threshold value such as 0.1 , so that the number of selected second users U ² is J, 1 j J , J I.

In step 56, the user terminal 1 obtains each second user according to the user data. An evaluation value of N _k relative to each of the historical events and the new event U ¹ and the first phase a user evaluation value for each of these historical events and the new event of the N _k To make the number of such historical events and the new events K, K 3,1 k K.

In this embodiment, each evaluation value By Department included in the user profile, and related to the first user U ¹ a browsing history for each of those articles of the user browsing behavior A _t of, for example, view the date and browse frequency and with the following Obtained by formula (1), each evaluation value By being included in the user profile and related to each second user For a user browsing history each of those articles of the browsing behavior of A _t of, for example, view the date and browse frequency and with the following formula (2) is obtained. In the example of Table 3, if the user U ¹ of the first user browsers such articles wherein one of A ₁ Browse date of the current date and the difference between 1 day and a frequency of 6 browser, and the user article A ₁ is classified as N ₁ , and the evaluation value of the first user U ¹ with respect to the event N ₁ That is If the first user U ^{1 does} not browse one of the user articles A ₂ , the R value and the F value are both 0, and the user article A ₂ is classified as N ₂ , then the first User U ¹ evaluation value relative to the event N ₂ That is 0.

R represents different score values corresponding to different browsing dates. When the difference between the browsing date and the current date is 0~3 days, the score value R is 4; when the difference between the browsing date and the current date is 4 days, the score value R is 3.2; When the difference between the browsing date and the current date is 5 days, the score value R is 2.4; when the difference between the browsing date and the current date is 6 days, the score value R is 1.2; when the difference between the browsing date and the current date is 7 Above day, the fractional value R is zero. F represents the browsing frequency, and the browsing frequency F is the ratio of the number of browsing times from the current date to the previous 6 days to 7. In this embodiment, the different scores R corresponding to different browsing dates, and the sampling times and sampling times corresponding to the browsing frequency may also vary according to requirements, and are not limited to the values disclosed in the embodiment.

In step 57, the evaluation values E ¹ , E ² obtained by the user terminal 1 according to step 56, and the first user U ¹ are relative to each second user. The browsing similarity S _j , capturing at least one of the historical events and the new event N as at least one recommended event corresponding to the first user U ¹ . In this embodiment, the usage end 1 is captured by a collaborative filtering algorithm.

It should be particularly noted that the collaborative filtering algorithm calculates a recommended value pref ( U ¹ , N _k ) of each of the historical events and the new event N by the following formula (3), and according to the The calculated recommended values pref ( U ¹ , N ) and the evaluation values E ^{1 are taken} from the historical events that are independent of the first user U ¹ and the recommended values are, for example, the first two highests and the new event N At least one of the preceding two heights is only an example, and in other embodiments of the present invention, the historical events with the highest recommended value or the top three high and at least the new event N may be taken. One, as the recommended event, is not limited to this.

For example, if each second user ,1 i a second user whose I has a browsing similarity S _i greater than the browsing threshold value, for example, 0.1 ,1 j J , J = 2 are respectively and And the first user U ^{1 is} opposite to the second user The browsing similarity S ₁ is 0.8, and the first user U ^{1 is} opposite to the second user. The browsing similarity S ₁ is 1, and the first user U ^{1 is} related to N ₁ , N ₃ , N ₄ , N ₅ , and the second user Relative to these historical events and each of the new events N _k , 1 k An evaluation value of K , K = 8 for =5, =3, =3, =0, =3, =0, =2 =0, the second user An evaluation value of N _k relative to each of the historical events and the new event for =0, =3, =2 =5, =0, =3, =0, =0, the recommended value of event N ₁ pref ( U ¹ , N ₁ ) is And so on, the recommended value of N ₂ is pref ( U ¹ , N ₂ ) is 3, the recommended value of N ₃ is pref ( U ¹ , N ₃ ) is 2.44, and the recommended value of N ₄ is pref ( U ¹ , N ₄ ) For 2.77, the recommended value of N ₅ is pref ( U ¹ , N ₅ ) is 1.33, the recommended value of N ₆ is pref ( U ¹ , N ₆ ) is 1.66, and the recommended value of N ₇ is pref ( U ¹ , N ₇ ) is 0.88. , N recommended value pref (U ^1, N ₈₎ of the _08, for example, so that the first two events is the high N ₂ and N ₆ of the first user U ¹ and independent recommended.

In step 58, the user terminal 1 includes a plurality of related message materials corresponding to each of the plurality of message users corresponding to each of the first news article and the second news article included in the recommended event. The vocabulary data of the negative vocabulary of the negative evaluation obtains a number of the plurality of message words included in each message material that meet the negative vocabulary.

In this embodiment, the user terminal 1 first uses the word breaking method related to the Chinese knowledge information processing, and the message material corresponding to each of the first news article and the second news article in the recommended event. The segments are divided into corresponding message words, and the number of the message words included in each message material that meets the negative words is obtained.

In step 59, the user terminal 1 obtains a ranking corresponding to each of the first news article and the second news article in the recommendation event according to the number corresponding to each message data, wherein the corresponding minimum is One of the first number of news articles and the second news article enjoys the first rank of the ranking, corresponding to one of the largest number of first news articles and second news articles enjoying the last order of the ranking. The user terminal 1 presents the presentation order of the first news article and the second news article in the recommendation event according to the ranking.

It should be particularly noted that the number of the message words included in each message material that meets the negative vocabulary can be used as a credibility indicating the first news article or the second news article corresponding to the message material. The reference of degree, when the number is larger, the lower the credibility of the first news article or the second news article corresponding to it; when the number is smaller, it represents the corresponding first news article or The higher the credibility of the second news article.

In summary, the first news article is obtained from each news server 2 by the user terminal 1, and each first news article is marked as one of the historical events and the new event. And obtaining, by the user terminal 1, the user data from the community server 3, and obtaining the browsing similarity S and the evaluation values E ¹ and E ² according to the user data, and further, according to the evaluation The values E ¹ , E ² and the browsing similarities S are used to recommend the recommended event to the first user U ¹ , and the recommended event recommended by the news tracking and recommendation method of the present invention is a personalized recommendation. Effectively reducing the time taken by the first user to search for news. In addition, the news recommended by the news tracking and recommendation method of the present invention is not a single news article but the recommended event, and the recommended event may include multiple articles related to each other. The news article, whereby the first user can obtain all relevant news articles at a time, thereby improving the convenience of the user to track the news, so that the object of the present invention can be achieved.

However, the above is only the embodiment of the present invention, and the scope of the invention is not limited thereto, and all the equivalent equivalent changes and modifications according to the scope of the patent application and the patent specification of the present invention are still The scope of the invention is covered.

51~59‧‧‧Steps

Claims (9)

A news tracking and recommendation method is implemented by a user connected to a plurality of news servers and a community server via a network. The news tracking and recommendation method comprises the following steps: (A) from each Corresponding to a news server of a news website, obtaining corresponding plurality of first news articles related to a first time interval; (B) corresponding to a second time interval preceding the first time interval according to the plurality of articles; And a second news article belonging to a plurality of different historical events, each of the first news articles being marked as one of the historical events and a new event; (C) corresponding to a first user U ¹ and the first plurality associated with the second user of the user U ¹ U ² of the server end of the community, to obtain user data associated with a user of the first user U ¹ and U ² of the second of these; (D) according to the user data to obtain the first user the second user U ¹ for each phase a browsing similarity S _i for the historical events and the new event N ; (E) selecting each second user according to each browsing similarity S _i a second user whose browsing similarity S _{i is} greater than a browsing threshold (F) obtaining each second user based on the user profile An evaluation value of N _k relative to each of the historical events and the new event U ¹ and the first phase a user evaluation value for each of these historical events and the new event of the N _k And (G) the evaluation values E ¹ , E ² obtained according to step (F) and the first user U ¹ relative to each second user The browsing similarity S _j captures at least one of the historical events and the new event N as at least one recommended event corresponding to the first user U ¹ . The news tracking and recommendation method according to claim 1, wherein in the step (A), the first news articles are obtained by using a web crawler technology. The news tracking and recommendation method according to claim 1, wherein the step (B) comprises the following sub-steps: (B-1) dividing each first news article into a corresponding plurality of first words according to a word-breaking method; a word, and dividing each second news article into a corresponding plurality of second words; (B-2) corresponding to the first word and each second news article corresponding to each first news article The second word uses a word frequency reverse file frequency method to obtain a plurality of first keywords corresponding to the first news article and a plurality of second keywords corresponding to the second news article; (B-3) Calculating a type of similarity of the same first news article relative to each second news article according to the first keywords corresponding to each first news article and the second keywords corresponding to each second news article And (B-4) marking the first news article as a historical event to which the second news article having the highest type similarity belongs and one of the new events according to the type similarity corresponding to each first news article By. The method of claim tracking and recommendation according to claim 3, wherein in step (B-1), each of the first news article and each second news article belongs to a Chinese language, and the word breaking method is A method of word breaking related to Chinese knowledge information processing. The news tracking and recommendation method according to claim 3, wherein in step (B-3), a cosine similarity formula is used to calculate the similarity of the same first news article relative to each second news article. degree. The method of claim tracking and recommendation according to claim 3, wherein the step (D) comprises the following sub-steps: (D-1) obtaining, according to the user profile, the first user U ¹ and the second a plurality of user articles of at least one of the users U ² ; and (D-2) dividing each user article into corresponding plurality of user words according to the word breaking method; (D-3) according to each The user words corresponding to a user article use the word frequency reverse file frequency method to obtain a plurality of user keywords corresponding to the user article; (D-4) corresponding to each user article Calculating the same user article relative to the first news article and the user keyword, the first keyword corresponding to each first news article, and the second keyword corresponding to each second news article An article similarity of each of the second news articles; (D-5) determining, according to the similarity of the articles corresponding to each user article, the similarity of the articles corresponding to the user article Whether one of them is greater than a threshold of use; (D-6) when the user article When one of the corresponding similarities of the articles is greater than the usage threshold, the user article is marked as one of the historical events and the new event N according to the similarity of the articles corresponding to the user article. Obtaining a mapping table of the first user U ¹ and the second user U ² with respect to the historical events and the new event N ; and (D-7) according to the mapping table Using a cosine similarity formula to obtain the first user U ¹ relative to each second user A browsing similarity S _i for the historical events and the new event N. The news tracking and recommendation method according to claim 6, wherein in step (F), each evaluation value Obtained by a browsing record included in the user profile and related to the browsing behavior of the first user U ¹ for each of the user articles, each evaluation value By being included in the user profile and related to each second user Obtained for a browsing record of the browsing behavior of each of these user articles. The news tracking and recommendation method according to claim 1, wherein in step (G), the recommended event is captured by a collaborative filtering algorithm. The news tracking and recommendation method according to claim 1, after step (G), further comprising the steps of: (H) according to each of the first news article and the second news article included in the recommendation event; Corresponding to a plurality of message materials related to the plurality of message users, and a vocabulary data comprising a plurality of negative words related to the negative evaluation, obtaining one of the plurality of message words included in each message material that meets the negative words And (I) obtaining, according to the number corresponding to each message material, a ranking corresponding to each of the first news article and the second news article in the recommendation event, wherein the minimum number of One of the news article and the second news article enjoys the first order of the ranking, corresponding to the maximum number Purpose One of the first news article and the second news article enjoys the last order of the ranking.