Data Mining News Headlines Classification Tool Computer Science Essay

Headline categorization Tool is a web application that can be used to sort intelligence headlines into different classs. Many of the bing intelligence categorization rely on full intelligence article content to sort a intelligence article into a specific class. We intend to construct an automatic categorization and analysis tool that uses merely the content in the intelligence headline for categorization. Headlines from different intelligence beginnings will be classified into assorted pre-defined classs utilizing supervised larning algorithms. We besides intend measure the effectivity of our tool by mensurating its truth.

Headline categorization tool is used to sort headlines into different classs. In order to execute this undertaking, we will foremost utilize a preparation informations set with content based on four different categories viz. - engineering, wellness, athleticss, and political relations. J48 and NaA?ve Bayes algorithms will be the two supervised larning algorithms that will be used to construct the classifier theoretical account.

After the theoretical account is trained utilizing either one of these algorithms, another set of informations will be used to prove the theoretical account and categorization will be generated.

Get quality help now
checked Verified writer
star star star star 4.7 (657)

“ Really polite, and a great writer! Task done as described and better, responded to all my questions promptly too! ”

avatar avatar avatar
+84 relevant experts are online
Hire writer

This tool will besides be used to measure the trial consequences and the truth of the tool. For rating, we can upload a individual headline or a file with multiple headlines, each of which can be of different class. The system will sort these headlines consequently into one of the four classs. It will supply a comparing between the category predicted by the user and the one predicted by the classifier.

Get to Know The Price Estimate For Your Paper
Number of pages
Email Invalid email

By clicking “Check Writers’ Offers”, you agree to our terms of service and privacy policy. We’ll occasionally send you promo and account related email

"You must agree to out terms of services and privacy policy"
Write my paper

You won’t be charged yet!

The category distribution of the categorization will be represented by a pie-chart generated utilizing Google Charts API. Finally, we will measure the classifier based on its truth, preciseness, and callback.

It is assumed that the user uploads relevant developing informations to construct the theoretical account and assigns the category for the preparation informations suitably. The tool provides really basic mistake managing mechanisms. Detailed mistake handling is beyond the range of this undertaking.

3. Requirement Specification

The undermentioned subdivision describes the functional demands of the system.

Shop Training Datas

The system shall let users to upload developing informations stored in a file. A preparation informations file can incorporate multiple headlines with each line in the file stand foring a individual intelligence headline. The system shall let the users to choose a category corresponding to the informations in the file. The system shall let the user to choose merely from one of the four pre-defined categories, viz. , Politicss, Sports, Technology, and Health. The system will hive away the preparation informations and its corresponding category in the database. This preparation informations will be used to construct the theoretical account that will be used for sorting the unobserved trial informations.

Generate Model

The system shall let coevals of the classifier theoretical account based on one of the two supervised larning algorithms. The two major categorization algorithms used by the system are J48 and NaA?ve Bayesian. Once the user selects desired algorithm, the system retrieves the preparation informations from the database and builds the theoretical account. The theoretical account and the preparation informations upon which it is built is once more stored in the database. It is of import to hive away the preparation informations matching to a theoretical account since any alteration in preparation informations will impact the theoretical account.

Classify Test Data

The system shall let users to sort a individual intelligence headline ( prove information ) into one of the four predefined classs. When the user inputs the intelligence headline, the system retrieves the latest generated theoretical account and its corresponding preparation informations from the database. The system uses this information to sort the unobserved trial case. The system besides calculates the category distribution to give the user an apprehension of how the trial information is classified. This category distribution shall be represented in the signifier of a pie-chart bespeaking the per centum distribution per category.

Model Evaluation

The system shall let users to measure the theoretical account by finding the theoretical account 's truth, preciseness and callback. This rating is performed utilizing bulk trial informations. The user uploads a file incorporating multiple intelligence headlines with each line of the file stand foring a individual headline ( test case ) . The system reads this file and so requests the user to foretell a category matching to each headline. Subsequently, the system besides determines the category of each trial case. The system so compares the category anticipations of user and the classifier to find the truth of the theoretical account. Class anticipations per category are besides recorded to cipher the preciseness and remember per category.

4. Categorization Algorithms and Pre-defined Classs

Supervised Learning Algorithms

Applications do non hold 'experiences ' . They learn from informations which are past experiences of application sphere. This undertaking is usually called supervised acquisition or inductive acquisition. In other words, in supervised acquisition, supervise or develop the system before, in order to acquire categorization of informations. Supervised acquisition is a sort of a machine larning undertaking in which preparation informations is used to bring forth a classifier. The preparation informations consists of an input object and a coveted category value. The supervised acquisition algorithm uses the information contained in the preparation informations to bring forth an `` inferred map '' besides known as a `` classifier '' . This classifier is so used to foretell the category value of unobserved trial cases.

In this undertaking, we use two supervised larning algorithms, viz. , J48 and Naive Bayes. J48 is an execution of C4.5decision tree larning which in bend green goodss determination tree theoretical accounts. In order to divide the possible anticipations it recursively splits a information set harmonizing to trials on property values. This algorithm uses avaricious attack to bring on determination trees for categorization. In this algorithm tree is constructed in a top-down recursive mode. All the preparation informations are at the root ab initio. Examples are partitioned recursively based on selected properties. As most of the supervised acquisition algorithms are built on analysing preparation informations and theoretical account is used to sort the trained informations, so is J48 algorithm. In this decision-tree theoretical account is built by analysing preparation informations and so theoretical account is used to sort trained informations. It generates determination trees and the node in the tree evaluates the being and significance of every single characteristic.

Naive Bayes algorithm is based on Bayesian theorem. It is sometimes besides termed as Probabilistic scholar. It uses all the properties contained in informations and so analyses each of them every bit as each of this case is independent of each other. Naive Bayes computes conditional chances of the categories mentioned in the case and so it picks the category which has got highest buttocks. Naive Bayes classifiers can be trained really expeditiously in a supervised acquisition manner. Naive Bayes categorization dainties each papers as `` bag of words '' . The theoretical account generated assumes words of a papers are generated independently of context given the category label. The chance of a word is independent of its place in the papers.

Pre-defined Classs for Categorization

There are batch of classs in any of the newspaper whether its paper or online. For this application we have picked up most the common classs which are wellness, political relations, engineering and athleticss. Sometimes some of intelligence web sites add subdivisions which are the most trending subjects for illustration in first hebdomad of November 2010, the most trending subdivision was election. So this subdivision was added some of the top intelligence web sites. In our undertaking we are non concentrating on these swerving subjects. It will be interesting to work with these categories as they are slightly related to each other. For case, on if you select the subdivision which says wellness, so on wellness class web page at the top it says 'Related Topic ' where engineering or health care classs are besides mentioned. As these classs are related to each other but tool is able to sort them right to a certain extent. That is why these classs will assist measure the theoretical account better.

5. System Architecture

Tomcat Servlet Container


Categorization Algorithm

Model Evaluation

Unobserved Instance


Training Datas


Browser- based Client

Classifier ( Model )

Figure 1: Headline Classification Application - System Architecture

The system consists of three-tier architecture. The browser is the user interface in the system. The user can upload the preparation informations, choose the categorization algorithm, and upload trial informations via the user interface. The web application, deployed in a servlet container, forms the nucleus of the system. The application contains the execution for parsing the preparation and trial informations and the categorization algorithms. The informations grade consists of the database used to hive away the preparation informations and theoretical account.

We use Apache Tomcat as the servlet container and MySQL as the database in the system.

6. Database Schema

The database for the application consists of two tabular arraies, viz. : TrainingData and Model

TrainingData tabular array scheme

Sample informations in TrainingData tabular array

Model table scheme

7. Execution

Headline categorization system is implemented as a web application. Users can entree this application from a web browser and the application itself is deployed in a servlet container. The application is developed utilizing Apache Struts, which is an unfastened beginning web application development model. The application chiefly uses WEKA Java APIs for all the undertakings related to pattern coevals and categorization. Specifically, WEKA APIs are used to:

Transform the natural preparation and trial informations into a standard format that can used for categorization.

Clean the preparation informations for halt words and instance transition.

Classification algorithm execution.

Model Evaluation.

The execution chief faculties of the web application are explained below and the corresponding categories in the beginning codification are besides indicated in bold text.

Training Data Storage ( )

Natural preparation informations in the signifier of apparent text file is uploaded by the user. The user besides provides the category to which the preparation informations belongs. This file contains multiple lines of intelligence headlines. The application parses each line in the file, and shops each line as a separate record in the database along with its associated category name.

Classification Model Generation ( )

Generating the theoretical account involves two major stairss:

Pre-process the preparation informations

The natural preparation informations can non be used as-is for categorization. This information has to be foremost transformed from natural text ( Nominal attributes ) to String informations type. This is done utilizing the WEKA 's NominalToString filter. Following, the String information has to be converted into a vector of numeral properties. This is done utilizing WEKA 's StringToWordVector filter. Before transition, the String information is converted to lowercase and halt words are removed to heighten the quality of preparation informations.

Generate the theoretical account

The theoretical account is generated utilizing WEKA APIs based on the categorization algorithm specified by the user. WEKA provides execution for J48 and Naive Bayes algorithm and we have used these algorithms to bring forth the theoretical account. Once the theoretical account is generated, the Byte watercourse of the theoretical account every bit good as a preparation informations matching to this theoretical account is stored in the database. This information is required during the Classification and Evaluation stairss.

Test Data Classification ( )

For sorting the unobserved trial case provided by the user, the theoretical account 's byte watercourse is foremost retrieved from the database and is cast back into a Classfier object. Similarly, the byte watercourse of the preparation informations used to construct the theoretical account is besides converted back into an Instance object. These two objects are so used to sort the trial case.

Model Evaluation ( )

To measure the theoretical account, trial informations is uploaded in majority. That is, a file incorporating multiple trial cases are uploaded by the user. It is necessary that the user be cognizant of the right category of each trial case when the user is prompted to foretell a category for each trial case. The system takes the input from the user, and classifies all the trial cases. A comparing is so drawn between the category anticipation made by the user and that made by the classifier. The system records the figure of duplicate anticipations. i.e the classifier predicted the same category as the user, every bit good as non-matching anticipations. These two statistics are used to cipher the truth of the theoretical account. Similarly, the system records anticipation informations for each category to find the preciseness and remember per category.

8. External Source Code Used

WebWeka hypertext transfer protocol: // # webweka

We reused the codification to preprocess and develop the theoretical account in the category

We besides reused to change over serialize and deserialize Model and Training Data objects in, and

WEKA API ( weka.jar )

Struts API ( struts.jar )

9. Input Data Collection

Two major types of informations are used for categorization - Training Datas and Test Data.

In order to accomplish maximal truth while sorting the trial information, it is necessary that the trial informations distribution be indistinguishable to that of the preparation informations.

Training Data Collection

News headlines across a span of two months were collected from hypertext transfer protocol: // A to develop the theoretical account. Four major classs of categorization viz. Health, Politics, Sports and Technology were identified and intelligence headlines related to all these Fieldss were stored in separate text files in the system. For developing the classifier, the text files were uploaded separately.

Test Data Collection

Current twenty-four hours intelligence headlines were collected as a portion of the trial informations.

10. Challenges Faced

Initially we proposed 'Automatic Twitter provender Classification and Analysis Tool ' as our undertaking. The aim of that undertaking was to pull out information from Twitter provenders and categorise those in pre-defined categories utilizing supervised larning algorithm. There are no tools available to categorise tweets into high degree subjects such as wellness, political relations, athleticss etc. We intended to construct such an automatic categorization and analysis tool that will supply an penetration into usage forms on Twitter. A The characteristics proposed for this undertaking were:

Classify tweets into assorted pre-defined classs utilizing a supervised acquisition algorithm

Determine the most tweeted class

Determine the most active users twirping about a peculiar subject

Show a ocular end product of the ascertained forms in a human apprehensible signifier.

Problems Encountered:

For making theoretical account utilizing developing informations, tweets from different classs were foremost collected as preparation informations. This dataset was converted to arff ( attribute relation file format ) format so that it can be used with unfastened beginning Java-based tool Weka. This arff file was fed to Weka tool utilizing filter as 'StringToWordVector ' with belongingss as

IDFTransform = True

lowerCaseTokens = True

Stopwords = ( need to advert it for Windows OS )

Tokenizer = AlphabetTokenizer

For the preparation set, theoretical account was generated successfully with right classification of categories. But when similar format of informations set was taken as trial set, so the categorization was non right. Initially when feeding trial informations in Weka, there was ever 'Test informations and preparation informations incompatible mistake ' encountered. This job was resolved by pre-processing trial informations and so utilizing it for categorization.

For categorization, different classifiers were used like Naive-Bayes, NaiveBayesMultinomial etc. But our major attempt for categorization was utilizing Naive-Bayes classifier. When utilizing trial informations, the categorization was non right. Every clip predicted category was different from the existent category.

We observed the ground for this was noisy informations, i.e. tweets are merely 140 characters long and consisted of particular characters like hash tickets ( for swerving subjects ) , and batch of stop words. In many instances, a URL was provided as a portion of a tweet without much meaningful information. We tried to take every bit much noise from the informations as possible. But in the terminal, we were left with really small informations that made sense. As a consequence, the theoretical account generated was non capable of accurately sorting the trial cases. We spent a batch of clip debugging this issue. We so came up with the thought of utilizing intelligence headlines which are besides short sentences like tweets, but most of the clip, they were complete sentences and made sense on their ain. We decided to recycle the application that we had built but merely change the preparation and trial informations. During the procedure of proving headline categorization, we observed that the categorization quality was up to the grade as headlines were more meaningful as compared to tweets.

Below is the sample tweet informations used for categorization:

Training Datas:

@ relation _Users_prachi_Documents_SJSU_Fall2010_CMPE296M_twitter_train_dataset

@ property text twine

@ attribute category { concern, athleticss }

@ informations

'Hoyer clyburn battle appears to be over. ' , concern

'Colorado Sen. Michael Bennet turns down DSCC chairmanship. ' , concern

'Sarah Palin 's unfavourable evaluation reaches an all clip high. ' , concern

'Sens. John Cornyn and Mark Warner, Reps. James Clyburn and Heath Shuler, Anita Dunn and Tom Davis. ' , concern

'Tiger one of four added to U.S. ' Ryder Cup roll. ' , athleticss

'Turkish squad paid UK recruit over $ 100K. ' , athleticss

' 's anticipations for the 2010 season. ' , athleticss

'Major conference saves leader Hoffman gets No. 600. ' , athleticss

Test Datas:

Test Data

@ property text twine

@ attribute category { concern, athleticss }

@ informations

'Sell-off on Wall Street - A sell-off in U.S. stocks picked up steam Friday afternoon, following a volatile trading s ' , concern

'Does Ford have a Cadillac scheme for Lincoln? - Ford Motor Company CEO Alan Mulally is carefully hedging the ' , concern

'Phillies win, regain NL East lead ' , athleticss

11. Classifier Evaluation

Classifier Evaluation helps us look into the rightness of the theoretical account built. The steps preciseness, callback and truth are used to measure the classifier.

Accuracy =

Preciseness =

Recall =

These prosodies are displayed in a tabular signifier for the user to look into the truth of the system. The classifier was evaluated with different preparation and trial sets.

Test Case



Small preparation informations set with 10 headlines in each class ( Sports, Health, Technology and Politics ) was chosen.

The trial set had two headlines under each class.

The classifier rating was less than 50 % accurate and 5 out of 8 trial sets were falsely classified.

The system learns merely from the preparation informations. So the preparation informations demands to be huge.

Training informations set with changing figure of informations in each class



Health -55

Technology - 40


While proving the application, headlines from unknown classs like Business, Education, etc gets classified under the class that has the maximal figure of developing sets.

( In A this instance -Health )

As categorization is done based on bag of words from the preparation set, the class in the preparation set with the maximal figure of headlines is chosen by default. This is a restriction of the classifier.

News headlines from similar classs.


Headlines from 'Science ' and 'Elections ' were added to the trial set.

Headlines from 'Science ' were classified under 'Technology ' and Headlines from 'Election ' were classified under 'Politics '

Choosing similar classs like 'Science ' and 'Technology ' or 'Politics ' and 'Election ' may ensue in incorrect anticipation by the classifier as the headlines in these classs overlap.

Each class in the preparation set had up to a 100 headlines.

The trial information was increased to ten per class

The consequences were better than when compared to utilizing few developing sets and with addition in developing informations, there was a relative addition in truth, preciseness and callback of the classifier.

With addition in developing informations, all the four classs ( Health, Politics, Technology and Sports ) which immensely differ from one another can be classified efficaciously.

Headlines that combine all the classs were used for proving to see how the classifier works.


Obama Gets Injured: Friendly Game Of Basketball Turns Into 12 Stitches.

Will the system sort it under Politics ( President ) , Sports ( Basketball ) or Health ( stitches )

The system classifies it into one of the three classs depending on the preparation informations set.

The system classifies the trial informations based on the count. The figure of times a word appears in the preparation informations in a peculiar class is used for categorization

12. Decision

The chief thought in the beginning was to make `` Automatic Twitter Feed Classification And Analysis Tool '' where we aimed at roll uping tweets from the Twitter web site and sorting them into different classs. After hebdomads of experimenting, we found that developing informations was the nucleus to constructing any categorization theoretical account. As the tweets in Twitter informations were really noisy, we were unable to accomplish good truths in the trial consequences.

A little fluctuation to the `` Twitter Feed Tool '' is the `` News Headlines Classification Tool '' where alternatively of roll uping noisy tweets, we collected headlines from different newspapers and aimed at sorting them into their several classs.

Even though a 100 per centum truth was non obtained, this undertaking aimed at sorting the top narratives in newspapers into their several classs based on headlines entirely. This undertaking gave us the apprehension of two algorithms ( J48 and Naive Bayesian ) . Experimenting with the algorithms utilizing the same preparation and trial sets helped us understand how otherwise they work. The truth of the theoretical account was mathematically displayed by ciphering the preciseness and callback. It besides gave us a really good apprehension of text excavation.

Testing with assorted preparation and trial sets helped us understand that the automatic categorization worked better with diverse classs which had different distinguishable words in their preparation informations ( headline ) therefore doing it easy for the theoretical account to foretell the class and accomplishing a higher truth

As a hereafter work, intense testing can be performed with different preparation and trial informations to understand the functionality of the different theoretical accounts and purpose at accomplishing better truths in sorting the headlines.

13. Application Screenshots

Upload preparation informations

Train Model

Classify Test Case

Evaluate Classifier

User Prediction

Classifier Prediction

Updated: May 19, 2021
Cite this page

Data Mining News Headlines Classification Tool Computer Science Essay. (2020, Jun 01). Retrieved from

Data Mining News Headlines Classification Tool Computer Science Essay essay
Live chat  with support 24/7

👋 Hi! I’m your smart assistant Amy!

Don’t know where to start? Type your requirements and I’ll connect you to an academic expert within 3 minutes.

get help with your assignment