TWI676110B - Semantic feature analysis system for article analysis based on readers - Google Patents

Abstract

The invention relates to a reader-centered semantic feature analysis system for article analysis, which includes a cloud server and more than one reader electronic device. The cloud server connects the reader electronic device through a network and receives one or more read Article information, the cloud server executes a semantic calculation process and a vocabulary calculation process according to the read article information to obtain a semantic vector information and a lexical vector information, respectively, and the cloud server according to the semantic vector information and the Vocabulary vector information, assessing the semantic features of the articles that the readers have read, and then providing readers with information about the articles of interest to the readers. Through the reader-centered semantic feature analysis, the articles are provided to the readers for reference, so as to improve the articles Purpose of analysis and recommendation accuracy.

Description

Semantic feature analysis system for reader-centered article analysis

The invention relates to an analysis system, especially a semantic feature analysis system that analyzes articles centered on readers.

With the rapid development of data information, each reader's acceptance and love of data information are different. These data information include newspapers, magazines, travel articles, shopping articles, consumer recommended articles and other articles, such as some readers I like to read shopping articles, and some readers like to read travel articles. Therefore, various shopping companies, travel companies, etc. all want to be able to recommend relevant article information about their own product information to suitable readers.

At present, most of the service methods for data information adopt unidirectional services. For example, various companies recommend article information to all readers, but not every reader is interested in these article information, which easily leads to reader rejection and poor service. situation.

In addition, there is another method of data information service. Each company collects the reading status of all article information through third-party systems such as Google Analytics, Facebook Pixal, short URL server, etc., and manually organizes and judges to confirm Which article information has a high reading rate (point collection rate), but this method is not only time-consuming, but also can not know the user's preferences for article information, so how to analyze the user's favorite article information from the perspective of readers, with a view to Providing more services is indeed an issue that needs to be improved.

In view of the problems existing in the foregoing prior art, the present invention provides a reader-centered semantic feature analysis system for article analysis. By receiving information about articles that the reader has read, the present invention analyzes the semantic features in order to grasp the reader's favorite features. Article information to provide readers with the most suitable article information.

A major technical means adopted in order to achieve the above purpose is a semantic feature analysis system for ordering the reader-centered article analysis, including: more than one reader electronic device for readers to read the article information; a cloud server, Connected to the reader electronic device via a network; wherein the cloud server receives the reader electronic device to return more than one read article information, and executes a semantic calculation program, the semantic calculation program performs the following steps: Read the article information and perform a word segmentation analysis to obtain a first word segmentation result; perform a high-dimensional vector calculation on the first word segmentation result according to a semantic vector algorithm to generate a semantic vector information.

According to the above content, after the cloud server receives the read article information and executes the semantic calculation process to understand the semantic vector information of the read article information that the reader has read, the cloud server can then The intention vector information provides suitable article information to the readers, and analyzes the semantic characteristics of the articles read by the reader as the center, so as to provide the corresponding articles for the readers' reference, so as to improve the accuracy of the article analysis and recommendation.

Another main technical means adopted to achieve the above purpose is to make the aforementioned semantic feature analysis system for reader-centered article analysis, including: more than one reader electronic device for readers to read the article information; a cloud server , And is connected to the reader electronic device via the network; wherein the cloud server receives the reader electronic device returning more than one read article information, and executes a vocabulary calculation program, the vocabulary calculation program performs the following steps: Read the article information for a word segmentation analysis to obtain a second word segmentation result; perform a one-dimensional vector calculation on the second word segmentation result according to a vocabulary vector algorithm to generate a vocabulary vector information.

According to the above content, after the cloud server receives the read article information and executes the vocabulary calculation process to understand the reader's reading of the vocabulary vector information of the read article information, the cloud server can then use the The vocabulary vector information provides suitable article information to the readers. By analyzing the semantic characteristics of the articles read by the reader as the center, the article is provided to the readers for reference, so as to improve the accuracy of article analysis and recommendation.

In order to achieve the above purpose, another major technical means is to make the aforementioned semantic feature analysis system for reader-centered article analysis, including: more than one reader electronic device for readers to read the article information; a cloud server Connected to the reader's electronic device via the network; wherein the cloud server receives the reader's electronic device to return more than one read article information, and executes a semantic calculation program and a vocabulary calculation program, the semantic calculation program executes the following Steps: Perform a word segmentation analysis on the information of the read article to obtain a first word segmentation result; perform a high-dimensional vector calculation on the first word segmentation result according to a semantic vector algorithm to generate a semantic intent The vocabulary calculation program performs the following steps: performing a word segmentation analysis on the read article information to obtain a second word segmentation result; performing a second word segmentation result according to a vocabulary vector algorithm One-dimensional vector calculation to generate a vocabulary vector information.

According to the above content, it is known that the cloud server receives the read article information and executes the semantic calculation process and the vocabulary calculation process to generate semantic vector information corresponding to the article information that the reader has read and use vocabulary vector information, thereby The reader-centered analysis of the semantic characteristics of the articles read by the readers, and accurately grasp the semantic characteristics of the information of the articles read by the readers, and accurately provide the articles suitable for the readers for the readers to read. The article conducts semantic feature analysis to provide readers with reference articles to achieve the purpose of improving the accuracy of article analysis and recommendation.

Regarding a preferred embodiment of a semantic feature analysis system for analyzing articles based on readers of the present invention, please refer to FIG. 1, which includes a cloud server 10 and more than one reader electronic device 20. The cloud server 10 is The network is connected to the reader electronic device 20. In this embodiment, the reader electronic devices 20 include electronic devices such as a personal computer, a notebook computer, a smart phone, a tablet computer, and a smart wearable device with networking, display, and operation functions.

The cloud server 10 includes a data analysis module 11, a data acquisition module 12, a data storage module 13, and a reading behavior feedback module 14. In this embodiment, a data reporting module 15 is further included. The data analysis module 11 is used to analyze the received data. The data acquisition module 12 is connected to more than one data database via the network to retrieve data and store it in the data storage module. 13. The data storage module 13 is used for storing data, and the reading behavior feedback module 14 is connected to the reader electronic device 20 via a network to receive one or more read article information returned by the reader electronic device 20, and the data report The module 15 is used to push information to the reader electronic device 20.

In this embodiment, the data database connected to the data extraction module 12 includes an academic thesis database, a seminar database, a news database, a dictionary database, a medical news database, etc. The database is just an example and is not limited here. By acquiring the data databases, the analysis accuracy of the data analysis module 11 is improved.

The main purpose of this system is to be reader-centered and analyze according to the information of each article read by the reader, in order to grasp the information of the article that the reader is interested in and like, and then provide the article that the reader may be interested in and like. In use, the reading behavior feedback module 14 receives the read article information returned by the reader electronic device 20 via the network, and sends the read article information to the data analysis module 11, which receives the Read the article information and execute a semantic calculation program and a vocabulary calculation program to analyze the read article information. For the steps of the data analysis module 11 to execute the semantic calculation program, please refer to FIG. 2. The semantic calculation program executes the following steps: performing word segmentation analysis on the information of the read article to obtain a first word segmentation result (S31); wherein the data analysis module 11 is based on the data storage module 13 The stored data database performs word segmentation analysis on the words of the read article information to delete the low-recognition word segmentation and optimize the A word segmentation result; a high-dimensional vector calculation is performed on the first word segmentation result according to a semantic vector algorithm to generate semantic vector information (S32); in this embodiment, the semantic vector algorithm is It is a word2vec-type neural algorithm. The word2vec-type neural algorithm is used to calculate the high-dimensional vector of the first word segmentation result. The high-dimensional vector calculation is performed with a dimension greater than 400 dimensions.

The data analysis module 11 executes the semantic calculation program, analyzes the received read article information, and obtains the semantic vector information, so that the semantic content of the article information read by the reader can be analyzed, and then Grasp the article information that readers are interested in reading.

In this embodiment, the semantic calculation program further includes the following steps: a standardized calculation is performed on the semantic vector information and the total number of the read article information to generate a standard value of the semantic vector (S33); In the embodiment, the semantic vector information is subjected to vector addition and divided by the total number of articles that have been read to generate a standard value of the semantic vector to quantify the semantic meaning of the article information for the company to quickly understand what the readers want. The semantic characteristics of the articles read.

In this embodiment, when a reader reads multiple articles within a set time, the reading behavior feedback module 14 receives the reader electronic device 20 to return information about multiple articles that have been read, and the data analysis module 11 The reading article information is organized into a summary of the read article information, and the data analysis module 11 executes the semantic calculation procedure on the aggregated read article information to generate corresponding semantic vector information, and further, the multiple read article information The corresponding semantic vector information also performs the normalization calculation, so as to divide the semantic vector information corresponding to the plurality of read article information by the total number of the read article information to obtain the corresponding semantic vector standard value. Among them, the way to organize the information of the read articles is to organize the information of the second read article at the end of the information of the first read article, so as to organize the information of the read articles in a sequential manner from head to end. .

In this embodiment, in order to evaluate whether the articles read by two or more readers are similar, the data analysis module 11 performs a vector inner product calculation on the semantic vector information of the articles read by the readers after analysis. Obtaining semantic vector similarity information, so that you can grasp the relevance of reading similar article information between different readers, so as to analyze the article information read by different readers in a large amount to match different readers in groups, which is convenient for inference. Broadcast information.

The data analysis module 11 executes the steps of the vocabulary calculation program. Please refer to FIG. 3, the vocabulary calculation program performs the following steps: perform a word segmentation analysis on the information of the read article to obtain a second word segmentation result ( S41); wherein, the data analysis module 11 performs a word segmentation analysis on the words of the read article information according to the data database stored in the data storage module 13 to delete the low-recognition word segmentation and obtain Optimized second word segmentation result; performing a dimensional vector calculation on the second word segmentation result according to a lexical vector algorithm to generate a lexical vector information (S42); in this embodiment, the semantic vector The algorithm is a tf-idf-type neural algorithm, and the tf-idf-type neural algorithm is used to calculate the dimension vector of the first word segmentation result. Among them, the dimension vector calculation performed by tf-idf is not limited in dimension, and The tf-idf-type neural algorithm can filter out repetitive and non-sense word segmentation, which is not included in the reference.

In this embodiment, the vocabulary calculation program further includes the following steps: A standardized calculation is performed on the total number of words of the used vocabulary vector information and the read article information to generate a standard value of the used vocabulary vector (S43); In the embodiment, the vector information of the vocabulary vector information is summed up and divided by the total number of words of the read article information to generate a standard value of the vocabulary vector information to quantify the words of the article information for the company to quickly understand the reader The word count of the article being read.

In this embodiment, after the reader has read multiple articles within a set time, the reading behavior feedback module 14 receives information about the multiple articles read by the reader's electronic device 20 within the set time, and the data is analyzed. Module 11 organizes the read article information into a summary of the read article information, and the data analysis module 11 executes the vocabulary calculation procedure on the collected read article information to generate corresponding vocabulary vector information, and further converts The vocabulary vector information corresponding to the read article information is also subjected to the normalized calculation, so as to divide the vocabulary vector information corresponding to the read article information by the total number of words of the read article information to obtain the corresponding vocabulary. Vector information.

In this embodiment, when evaluating whether the articles read by two or more readers are similar, the data analysis module 11 performs a vector inner product calculation using the vocabulary vector information after analyzing the articles read by the readers. Obtain a similarity information using vocabulary vectors, so that you can grasp the relevance of reading similar article information between different readers. By analyzing a large number of vocabularies of article information read by different readers, you can distinguish the differences in reading relevance. Readers are grouped and matched to facilitate the dissemination of information.

After the system analyzes the article information read by the reader, the data analysis module 11 generates more than one recommended article information according to the semantic vector information and the vocabulary vector information, and uses the data reporting module 15 to The recommended article information is broadcasted to these reader electronic devices 20 through a social media, a web page, an APP, an email, and a digital video playback media for different readers to refer to the article information of interest.

In this embodiment, the social media includes various social media such as LINE, Facebook, WeChat, Whatsapp, Weibo, Instagram, Twitter, Snapchat and so on.

In this embodiment, the webpage includes webpages such as http, html, php, asp, and jsp.

In this embodiment, the digital video playback media includes Youtube, netflex, iQiyi, etc. Digital video playback media that can broadcast text messages during video playback.

In this embodiment, the above-mentioned recommended article information promoted by social media, web pages, APPs, emails, digital audiovisual media, etc., can be displayed through these reader electronic devices 20 through two-dimensional bar codes and web page links.

Please refer to FIG. 4 for an example of LINE. Through the data reporting module 15, the recommended article information that may be of interest to the reader is pushed to the reader ’s electronic device 20 ’s LINE for the reader ’s reference and the web page is displayed. The link is for readers to click on the corresponding webpage to read the recommended article information.

10‧‧‧ Cloud Server

11‧‧‧Data Analysis Module

12‧‧‧Data Acquisition Module

13‧‧‧Data Storage Module

14‧‧‧Reading as feedback module

15‧‧‧ Data Report Module

20‧‧‧ Reader Electronic Device

FIG. 1 is a block diagram of a system architecture according to a preferred embodiment of the present invention. FIG. 2 is a flowchart of a semantic calculation program according to a preferred embodiment of the present invention. FIG. 3 is a flowchart of a vocabulary calculation program according to a preferred embodiment of the present invention. FIG. 4 is a schematic diagram of providing article information according to a preferred embodiment of the present invention.

Claims (8)

A reader-centered semantic feature analysis system for article analysis includes: more than one reader electronic device for readers to read article information; a cloud server connected to the reader electronic device via the network and via the network Connect to more than one data database to retrieve data; wherein the cloud server receives the reader's electronic device to return more than one read article information, and executes a semantic calculation program, which performs the following steps : Perform a word segmentation analysis on the information of the read article to obtain a first word segmentation result; perform a high-dimensional vector calculation on the first word segmentation result according to a semantic vector algorithm to generate a semantic vector Information; and a standardized calculation of the semantic vector information and the total number of articles that have been read to generate a standard value of the semantic vector, where the semantic vector information is summed by the vector and divided by the read The total number of article information to generate the standard value of the semantic vector to quantify the semantic meaning of the article information. The reader-centered semantic analysis system for article analysis as described in claim 1, wherein: when the cloud server receives information about multiple articles that have been read, the information about the articles that have been read is compiled into a collection of articles that have already been read Information, and execute the semantic calculation process for the aggregated read article information to generate corresponding semantic vector information; wherein the cloud server performs the standardized calculation on the semantic vector data generated by the aggregated read article information, To generate the corresponding standard value of the semantic vector. The reader-centered semantic analysis system for article analysis as described in claim 2, wherein: when the cloud server receives the read article information returned by different reader electronic devices, the different reader devices and the read article information are compared. The semantic calculation program is executed to generate corresponding semantic vector information, and the semantic vector information of different reader devices is calculated by a vector inner product method to obtain the semantic vector similarity information. A reader-centered semantic feature analysis system for article analysis includes: more than one reader electronic device for readers to read article information; a cloud server connected to the reader electronic device via the network and via the network Connect to more than one data database to retrieve data; the cloud server receives the reader's electronic device to return more than one read article information, and executes a vocabulary calculation program, which performs the following steps : Perform a word segmentation analysis on the read article information to obtain a second word segmentation result; perform a one-dimensional vector calculation on the second word segmentation result based on a vocabulary vector algorithm to generate a vocabulary vector information ; And performing a normalized calculation of the total word count of the used vocabulary vector information and the read article information to generate a standard value of the used vocabulary vector, where the vector sum of the used vocabulary vector information is divided by the read article The total number of words of the information to generate the standard value of the vocabulary vector used to convert the article information Words quantified. The reader-centered semantic analysis system for article analysis as described in claim 4, wherein: when the cloud server receives information about multiple articles that have been read, the information about the articles that have been read is compiled into a collection of articles that have already been read Information, and execute the vocabulary calculation process for the aggregated read article information to generate corresponding vocabulary vector information; wherein the cloud server performs the standardized calculation on the vocabulary vector information generated from the aggregated read article information, To generate the corresponding standard value of the vocabulary vector. The reader-centered semantic analysis system for article analysis as described in claim 5, wherein: when the cloud server receives the read article information returned by different reader electronic devices, the reader device and the read article information are compared. The vocabulary calculation program is executed to generate corresponding used vocabulary vector information, and the used vocabulary vector information of different reader devices is calculated by a vector inner product method to obtain a used vocabulary vector similarity information. A reader-centered semantic feature analysis system for article analysis includes: more than one reader electronic device for readers to read article information; a cloud server connected to the reader electronic device via the network and via the network Connect to more than one data database to retrieve data; wherein the cloud server receives the reader's electronic device to return more than one read article information, and executes a semantic calculation procedure and a vocabulary calculation procedure, the semantic meaning The calculation program performs the following steps: performing a word segmentation analysis on the read article information to obtain a first word segmentation result; performing a high-dimensional vector calculation on the first word segmentation result according to a semantic vector algorithm, and Generating semantic vector information; and performing a normalized calculation of the semantic vector information and the total number of articles read to generate a standard value of the semantic vector, wherein the semantic vector information is summed into a vector Divide by the total number of articles that have been read to generate the standard value of the semantic vector to Quantification; where the vocabulary calculation program performs the following steps: perform a word segmentation analysis on the read article information to obtain a second word segmentation result; perform a second word segmentation result based on a vocabulary vector algorithm Dimension vector calculation to generate a vocabulary vector information; and a standardized calculation of the total vocabulary vector information and the total number of words of the read article information to generate a vocabulary vector standard value, where the vocabulary vector is used The information is summed up by vectors and divided by the total number of words of the read article information to generate a standard value of the vocabulary vector used to quantify the words of the article information. The reader-centered semantic feature analysis system as described in claim 7, wherein: the cloud server generates one or more recommended article information according to the semantic vector information and the vocabulary vector information, and Social media, a webpage, an APP, an email, or a digital video playback media are broadcast to these reader electronic devices.