Volume 5, Issue 10, October – 2020 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Hybrid News Recommendation System using TF-IDF

and Machine Learning Approach
C.P. Patidar, Dr. Meena Sharma, Yogesh Katara
Department of Information Technology,
IET DAVV, Indore M.P., India

Abstract:- A newspaper divided into various sections like concept to evaluate and sort content according to demand
a city, sports, editorial, international, national, and popularity. Generally, it observes the description
entertainment etc. All the above sections have equal associated with items or existing content and compares
importance and different user followers for different with user preference.
sections. Sometimes there may be a possibility that they 2. Collaborative separating frameworks prescribe things
may consist of relevant information but in different dependent on comparability criterion among clients or
sections and different newspapers. News potentially things. The things prescribed to a client are
Recommendation System can overcome this problem those favoured by comparable clients. This kind of
and suggest relevant news according to user preference recommendation can utilize the preparation on likeness
and popularity factor. This research paper investigates search and bunching. Nonetheless, these innovations
the need for news recommendation using a machine without anyone else's input are not adequate, and some
learning approach to make it more efficient and better. new calculations that have demonstrated compelling for
Hybrid Approach can help to recommend news to users recommendation system.
based on Supervised Machine Learning and Term 3. Knowledge frameworks prescribe proposals or
Frequency-Inverse Document Frequency (TF-IDF). arrangement by producing physically or naturally various
ends and choice standards. It stresses on express field
Keywords:- Machine Learning, Naïve Bayes, News information about the necessities and client inclination. On
Recommen- Dation, TF-IDF. the other hand, physically created choice principles or
made inferences might be one-sided and not appropriate
I. INTRODUCTION for customized frameworks. This framework related to
various downsides, for example, bottleneck issue during
Tosearch for desired information user faces difficulty information handling and acquire issue during client
because of getting irrelevant information. This issue arises profile creation and connecting with existing data. A
due to the insufficient knowledge of search tools and programmed information-based framework is prescribed
availability of a large amount of data. Extraction of desired where the contribution of information might be emotional
information becomes difficult in this case. The and can fluctuate as indicated by prerequisite.
recommendation system is beneficial in this case, which 4. A demographic recommender system gives suggestion
offers a relevant set of information. The examination dependent on a client's statistic profile, which includes the
analyzes, a broad application or device that includes client client's statistic information, for example, sex, age, date of
inclination or self-gathered information for predicting birth, instruction and other individual highlights. This
client’s need and investigates the best probability of methodology classifies clients into bunches dependent on
importance among data which is known as recommendation their statistic attributes and suggests protests as needs are.
System, or it tends to be express that recommendation system All the more definitely, it accept clients in a similar
is the device that gives pre-determined information-based classification have a similar taste or inclinations. Proposals
data. Recommendation system may valuable in different are given for new clients by first recognizing the
fields, for example, news, shopping, item search and so on. classification client have a place with and afterwards by
finding inclinations of different clients in a similar class.
The recommendation system utilizes various
advancements. Recommendation system is classified as: II. RELATED WORK
1. Content-based recommender system works on user
preference and content exists at the data source. It Different recommendation system techniques and their
compares and extracts the information from web pages pitfalls are mentioned in table1.
and data sources and matches with user preference. It also
uses popularity calculations and frequent uses to find the
most used and most demanding content. It uses this

IJISRT20OCT630 www.ijisrt.com 1033

Volume 5, Issue 10, October – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
TABLE 1
LITERATURE REVIEW

S. Title Tools/Techniques Concept Advantage Disadvantage

No

1. Cold Start KNN, Combines with A significant Cold start

Recommenda- Collaborative the attribute improvement in problem.
tion Based on Filtering, Matrix information of recommendation
Attribute-Fused Factorization. the item with accuracy with
Singular Value the historical solving cold start
Decomposition rating matrix to problem of new
[1]. predict the items.
potential
preferences of
the user.

2. A Recommendation Proposed a new Improves the It does not

Recommendatio model based on distance to accuracy of the optimize the
n Model Based content and social calculate the recommendation, similarity of
on Content and network text similarity reduces the cost text and
Social Network (RMBCS). between long and cost of improve
[2]. text and short training, also recommended
text. After that enhances the performance.
nearest novelty of the
neighbour recommendation.
group is found
from user’s
social network.
Then,
recommend the
texts.

3. A New K-means It proposes and Significantly Scalability is

Collaborative algorithm and evaluates an improved the not achieved.
Filtering Singular Value effective two performance of
Recommendatio- Decomposition stage recommendations
n Algorithm (SVD). recommender and remained the
Based on system that can lowest values in the
Dimensionality generate RMSE curve in the
Reduction and accurate and whole neighbours
Clustering highly efficient range.
Techniques [3]. recommend-
ations.

IJISRT20OCT630 www.ijisrt.com 1034

Volume 5, Issue 10, October – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Y. Ma, et al. [4] described that in group-oriented well as based on item ranking techniques that can generate
recommendation field, the design of a commonly acceptable more fine results.
recommendation list is a tough task. Traditional group
recommendation algorithms those are used in the III. PROBLEM DOMAİN AND OBJECTİVES
recommendation are often realize group recommendation list
aggregation according to the item ranking or their item score News reading is one of the most common activity of
of group members’ recommendation lists. The factors that daily life. The developing web world and rushed timetable of
are considered in these algorithm is relatively one-sided. day-by-day life make such a great amount of trouble for web
users to discover related news. This circumstance turns out to
L. Zonglei, et al. [5] demonstrated a new method to be more regrettable when client attempt to inquire data and
forecast flight delays. This new method is based on the get immaterial news content. Insufficient learning of pursuit
content-based recommendation system. In the forecast machine and an extensive measure of information gives poor
model, the events such as flight delays and airportstn that execution to recover or separate news content. Suggestion
have been mapped to users and items, respectively, which are frameworks offer scholarly practice in view of client
the concepts in the recommendation system. According to inclination. Proposal frameworks offer a discrete and specific
the propagation of delay, this method alerts the target arrangement of data. As of late, Web personalization for
airport by monitoring the status of the related airports. The news has gotten much regard for help Internet clients with
observed status is compared with the historical data to the issue of data over-burden.
predict the seriousness of delay. Since the airborne hours
between every two airports are usually more than an hour, Following points are expected from proposed research
this method could give the alarm at least 1 hour ahead. work:
Besides, the above factors the method requires minimum  To load and clean data of BBC news dataset and load for
online calculation, and therefore it guarantees that the delay lemmatization and filtering.
forecast can be delivered in a quick and timely manner.  To implement the Naïve Byes classification algorithm to
classify the data into multiple categories.
Bahram amini, et al. [7] focuses on user search in a  To implement TF-IDF algorithm and recommend news
recommendation system. User profile plays a major role articles accordingly.
infiltration techniques as user profile signifies what one can  To recommend news articles based on classification and
search. User logs are a wide collection of data hence searches TF-IDF algorithm along with estimate results based on
should be specific. This study gives a brief overview of a accuracy, precision and f-score.
recommender system. Data from different sources which are
searched is considered in this work. Personalization system is IV. PROPOSED WORK
classified in several ways some are utility functions or call
modelling. These work further describe a hybrid approach The news recommendation system is used to have the
which combines content-based, collaborative based and desired information while searching. Different news content
knowledge-based approach. The knowledge-based system may have different news category. Sometime, the news
generates recommendation with the help of decision rules. category may be known before recommendation but
All the specifications of users are analyzed then knowledge- sometimes no one knows about news category. We have to
intensive rules that are generated based on users choice use a learning approach to identify the category of news and
having similarities. Traditional content-based recommend them according to relevancy factor. A Hybrid
recommendation uses the data on web pages and ratings of it Solution using machine learning-based Naïve Bayes
which user browsed. The comparison of user profile and data classification technique along with TF-IDF algorithm has
on the web pages are done. The traditional scheme follows been proposed to make a common and most relevant
the pattern in which it strongly believes that new choices are recommendation. A block diagram to explore this solution is
highly influenced by the past. depicted in Fig 1.
Adomavicius, et al. [8] address that recommender
systems are becoming increasingly important to individual
users and businesses for providing personalized
recommendations. They investigate that most of the
researchers have only focused on recommendation accuracy,
other important aspects of recommendation quality, such as
the diversity of recommendations, have often been
overlooked. It also suggests that the recommendation system
are highly important in the current world scenario as data on
browsers is very huge. Individuals, as well as business, need
a class level of recommenders. Investigations observed that
most of the researchers have focused on the accuracy of
recommendation, quality and diversity are mostly ignored. In
this paper, the recommendation is given based on accuracy as
Fig 1:- Flow of the proposed recommendation system

IJISRT20OCT630 www.ijisrt.com 1035

Volume 5, Issue 10, October – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The complete work has been classified into four  Calculation of IDF [Inverse Document Frequency]
modules which are following: Equation (2) shows the IDF weight of term t.

Module 1: Dataset idft = log10

N
(2)
BBC Dataset has been recommended to consider as the dft
input source for news recommendation.
Important Points:
Module 2: Classification using Machine Learning  Wher N is the number of documents in the collection.
Machine learning is used to first learn the concept and  idft is a measure of the informativeness preset in the
then apply intelligence to take a decision. This module will document of the term.
help to learn about the thought behind categorization of news
and relevance to categorize unlabeled news into categories.  Calculation of TF-IDF weighting
Here, the Naïve Bayes classification algorithm has been used Equation (3) shows the tf-tdf weight of a term is the
to classify the data into multiple categories. Initially, a product of its tf weight and its idf weight.
training step will be used to provide learning to data then 𝑁
work will be classified according to learn thought and during 𝑊𝑡,𝑑 = (1 + log10 𝑡𝑓𝑡,𝑑 ). log10 (3)
𝑑𝑓𝑡
the testing module. Then after user desired category news
will be forwarded to the next module for the top Outcome of this module generates the product sum for
recommendation. every document which will help to evaluate top ten News
recommendation.
Module 3: TF-IDF Algorithm
TF-IDF is an information retrieval(IR) algorithm based Module 4: Similarity Matching & Recommendation
on the occurrence of keywords in the whole dataset as well This module will take user choice in terms of keywords
as particular documents. A detailed description of the DF and threshold value to decide the cutoff for the
calculation is cited below; recommendation. A final bunch of news will be
recommended as a final output.
 Calculation of Document Frequency
 The term frequency tft, d of term t in the document d can V. CONCLUSİON
be defined as the number of times that t occurs in d.
This research work addresses the need of modern news
 A document with the tf = 10 occurrences of the term is
recommendation system based on user choice. This research
more relevant than a document with the
work identifies that machine learning approach could help to
 tf = 1 occurrence of the term.
classify the news into multiple categories and TF-IDF could
 Relevance does not increase proportionally with term help to find the similarity factor and decide most relevant
frequency. news. A model of the proposed solution is also developed
 The document frequency can be defined as the number of and define inside proposed work. This work will be
documents in a collection that the term occurs in. implemented using java technology and it will be evaluated
 dft is the document frequency can be defined as the based on precision, recall and f-score along with computation
number of documents that t occurs in. time to measure computation performance.
 dft can be defined as an inverse measure of the
informativeness of term t. ACKNOWLEDGMENT
 Calculation of Log frequency weighting This paper and the research behind it would not have
Following steps are executed to calculate log frequency been possible without the exceptional support of my
weighting: supervisor, Dr. Meena Sharma and Mr. C.P.Patidar. His
 Equation (1) shows the log frequency weight of term t in enthusiasm, knowledge and attention to detail have been an
d. inspiration and kept my work on track.

1 + log10 tft,d iftft,d > 0 REFERENCES

Wt,d = { (1)
0 otherwise
[1]. X. Guo, S. Yin, Y. Zhang, W. Li and Q. He, "Cold Start
 Score for a document-query pair can be given as a sum Recommendation Based on Attribute-Fused Singular Value
over terms t in both qand d: Decomposition," in IEEE Access, vol. 7, pp. 11349-11359,
2019.
Tf-matching-score(q, d) can be calculated as: [2]. H. Xue and D. Zhang, "A Recommendation Model Based
t∈q∩d (1 + log tft,d ) on Content and Social Network," 2019 IEEE 8th Joint
 The score is said to be 0 if none of the query terms are International Information Technology and Artificial
present in the document. Intelligence Conference (ITAIC), Chongqing, China, 2019,
pp. 477-481.
[3]. H. Zarzour, Z. Al-Sharif, M. Al-Ayyoub and Y. Jararweh,
"A new collaborative filtering recommendation algorithm

IJISRT20OCT630 www.ijisrt.com 1036

Volume 5, Issue 10, October – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
based on dimensionality reduction and clustering
techniques," 2018 9th International Conference on
Information and Communication Systems (ICICS), Irbid,
2018, pp. 102-106.
[4]. Y. Ma, S. Ji, Y. Liang, J. Zhao and Y. Cui, "A Hybrid
Recommendation List Aggregation Algorithm for Group
Recommendation," 2015 IEEE/WIC/ACM International
Conference on Web Intelligence and Intelligent Agent
Technology (WI-IAT), Singapore, 2015, pp. 405-408.
[5]. L. Zonglei, W. Jiandong and X. Tao, "A new method for
flight delays forecast based on the recommendation system,"
2009 ISECS International Colloquium on Computing,
Communication, Control, and Management, Sanya, 2009,
pp. 46-49.
[6]. Barskar, N. and Patidar, C.P., 2016. A survey on cross
browser inconsistencies in web application. International
Journal of Computer Applications, 137(4), pp.37-41.
[7]. Bahram amini, roliana ibrahim, mohd shahizan othman,
"Discovering the impact of knowledge in recommender
systems: a comparative study", International Journal of
Computer Science & Engineering Survey, vol 2,pp-3,2011.
[8]. Adomavicius, G, & Kwon, Y. O. (2012). Improving
aggregate recommendation diversity using ranking-based
techniques. IEEE Transactions on Knowledge and Data
Engineering, 24(5), 896-911.
[9]. Patidar CP, Sharma M. An Automated Approach for Cross-
Browser Inconsistency (XBI) Detection. InProceedings of
the 9th Annual ACM India Conference 2016 Oct 21 (pp.
141-145). ACM.
[10]. Aizawa, A., 2003. An information-theoretic perspective of
tf–idf measures. Information Processing & Management,
39(1), pp.45-65.
[11]. J. Liu, M. Tang, Z. Zheng, X. F. Liu, and S. Lyu, "Location
aware and personalized collaborative filtering for web
service recommendation," IEEE Transactions on Services
Computing, vol. 9, no. 5, pp. 686-699, 2016.
[12]. J. Wang, A. P. De Vries, and M. J. Reinders, "Unifying user
based and item-based collaborative filtering approaches by
similarity fusion," in Proceedings of the 29th annual
international ACM SIGIR conference on Research and
development in information retrieval, 2006, pp. 501-508.
[13]. Patidar, C., Sharma, M. and Sharda, V., 2017. Detection of
cross browser inconsistency by comparing extracted
attributes. International Journal of Scientific Research in
Computer Science and Engineering, 5(1), pp.1-6.
[14]. M. Aleksandrova, A. Brun, A. Boyer, and O. Chertov,
"Identifying representative users in matrix factorization-
based recommender systems: application to solving the
content-less new item cold-start problem," Journal of
Intelligent Information Systems, vol. 48, no. 2, pp. 365-397,
2017.
[15]. S. Deng, L. Huang, G. Xu, X. Wu, and Z. Wu, "On deep
learning for trust-aware recommendations in social
networks," IEEE transactions on neural networks and
learning systems, vol. 28, no. 5, pp. 1164-1177, 2017.

IJISRT20OCT630 www.ijisrt.com 1037