Aspect Level Information Retrieval System for Micro Blogging Site
1 like316 views
The paper presents a method for aspect-level information retrieval by calculating the polarity of public tweets on various current affairs topics, including fashion, crime, safety, corruption, and inflation. Using positive and negative bags of words, the study evaluates the sentiment expressed in these tweets, revealing overall negative sentiment towards fashion, crime, and inflation, while showing higher positivity for safety. The research aims to provide insights into public opinions and can continuously process data without privacy concerns.
![International Journal of Science and Research (IJSR), India Online ISSN: 2319‐7064
Volume 2 Issue 8, August 2013
www.ijsr.net
Aspect Level Information Retrieval System for
Micro Blogging Site
Manisha Punn1
, Nidhi Bhalla2
1, 2
Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India
Abstract: In this paper we have calculated the polarity of people on two current affairs. For this, the tweets posted by twitter handler
network friends and friends of friends on current affairs are extracted from twitter. Following this, two bags of words are created. One is
positive bag of words and other is negative bag of words. The objective aspect based polarity evaluation is implemented. The polarity of
tweets have illustrated and interpreted with application to understand current digital word in terms of its aspect, perspective by
calculating polarity with parameter influencing thoughts, concepts, ideas and aspects of people expressing in digital world
Keywords: Aspect, Information retrieval, Opinion mining, Perspective, Tweets
1. Introduction
Information retrieval deals with acquiring relevant
information from a collection of information resources. After
retrieving the information about web objects, the information
is indexed and then ranked. Aspect[1] is a related object or
member or part of a topic of interest on which the evaluation
is made. Aspect level information retrieval system retrieves
the information based on his/her aspect, perspective etc.
The present paper retrieves aspects of five celebrities on five
different topics and using them proposes an algorithm for
calculating the polarity of users’ aspects. The polarity is
calculated as either positive polarity or negative polarity.
2. Previous Work
In a research work carried by [2], a novel unsupervised
model known as Cross-Perspective topic model was
proposed for contrastive opinion mining. The opinions on
any ad hoc query were determined based on learned
parameters. This model quantified the difference in user
perspectives. The research was carried out on two datasets,
that is, statement records of U.S senators and world news
report from three media in U.S, China and India. It is
concluded that the proposed model showed effective results.
Another study performed by [3] proposed a feature wise
opinion mining system which extracted features from user
generated contents and then determined the intensity of those
opinions. The positive and negative features were identified
by extracting the associated modifiers and opinions. It was
concluded that the recall value was lower than precision
which means that certain correct feature-opinion pairs could
not be recognized by the system correctly. Also, more
number of features could not be extracted in negative list as
most of the reviews written by users were on positive side.
3. Methodology of Proposed Work
Following steps were performed in order to carry out the
proposed work as follows:
Figure 1: Methodology of proposed work
3.1. Select top 5 current subjects
The first step is to select the top five current subjects. The
subjects were selected using Google Trends [4]. Google
Trends show the total search volume of a particular search
term across the world.
359](https://image.slidesharecdn.com/mtiwmtmxmtk-131002084318-phpapp01/85/Aspect-Level-Information-Retrieval-System-for-Micro-Blogging-Site-1-320.jpg)
![International Journal of Science and Research (IJSR), India Online ISSN: 2319‐7064
Volume 2 Issue 8, August 2013
www.ijsr.net
is no breach in privacy as only public dataset of tweets is
processed, and it can run continuously to bring the statistical
information , what percentage of people are thinking in what
way-positive or negative. It is from this views we can access
the content of thoughts and their flow of attributes, feelings,
mood direction of the people towards the current affairs
topics like fashion, crime, safety, corruption and inflation.
Future Scope
In our current research we have tried to capture the views,
expressions and thoughts of people using digital tools like
twitter and have formulated mathematical expressions for
calculating their positive aspects and negative aspects of
their expressions (tweets) in an non-invasive way, which
produced fairly good realistic view of how people are
thinking on current aspects of things happening around them,
for future scope however we suggest a deeper analysis may
be conducted to extend this work. The further analysis may
consist of identifying trends between various aspects of
people’s opinions with identifying causal relationships
between influencing parameters. This may be conducted
using regression also.
References
[1] Nozomi Kobayashi, Kentaro Inui, and Yuji Matsumoto,
“Extracting Aspect-Evaluation and Aspect-of Relations in
Opinion Mining”, 2007.
[2] Yi Fang, Luo Si, Naveen Somasundaram, Zhengtao Yu,
“Mining Contrastive Opinions on Political Text Using
Cross-Perspective Topic Model’, 2012.
[3] Tanvir Ahmad, Mohammad NajmudDoja, “Ranking
Sysem for Opinion Mining of Features from Review
Documents”, 2012.
[4] http://www.google.com/trends/
Authors Profile
Manisha Punn is pursuing M. Tech (Computer
Science) form Swami Vivekanand Institute of
Engineering and Technology, Banur, Punjab, India.
She did her B. Tech (Information Technology) form
Doaba Institute of Engineering and Technology, Kharar, Punjab,
India.
Nidhi Bhalla is working as Assistant Professor in
Swami Vivekanand Institute of Engineering and
Technology, Banur, Punjab, India. She did her B. Tech
(Information Technology) and M. Tech (Computer
Science) form Lovely Professional University, Jalandhar, Punjab,
India.
362](https://image.slidesharecdn.com/mtiwmtmxmtk-131002084318-phpapp01/85/Aspect-Level-Information-Retrieval-System-for-Micro-Blogging-Site-4-320.jpg)
Aspect Level Information Retrieval System for Micro Blogging Site
- 1. International Journal of Science and Research (IJSR), India Online ISSN: 2319‐7064 Volume 2 Issue 8, August 2013 www.ijsr.net Aspect Level Information Retrieval System for Micro Blogging Site Manisha Punn1 , Nidhi Bhalla2 1, 2 Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India Abstract: In this paper we have calculated the polarity of people on two current affairs. For this, the tweets posted by twitter handler network friends and friends of friends on current affairs are extracted from twitter. Following this, two bags of words are created. One is positive bag of words and other is negative bag of words. The objective aspect based polarity evaluation is implemented. The polarity of tweets have illustrated and interpreted with application to understand current digital word in terms of its aspect, perspective by calculating polarity with parameter influencing thoughts, concepts, ideas and aspects of people expressing in digital world Keywords: Aspect, Information retrieval, Opinion mining, Perspective, Tweets 1. Introduction Information retrieval deals with acquiring relevant information from a collection of information resources. After retrieving the information about web objects, the information is indexed and then ranked. Aspect[1] is a related object or member or part of a topic of interest on which the evaluation is made. Aspect level information retrieval system retrieves the information based on his/her aspect, perspective etc. The present paper retrieves aspects of five celebrities on five different topics and using them proposes an algorithm for calculating the polarity of users’ aspects. The polarity is calculated as either positive polarity or negative polarity. 2. Previous Work In a research work carried by [2], a novel unsupervised model known as Cross-Perspective topic model was proposed for contrastive opinion mining. The opinions on any ad hoc query were determined based on learned parameters. This model quantified the difference in user perspectives. The research was carried out on two datasets, that is, statement records of U.S senators and world news report from three media in U.S, China and India. It is concluded that the proposed model showed effective results. Another study performed by [3] proposed a feature wise opinion mining system which extracted features from user generated contents and then determined the intensity of those opinions. The positive and negative features were identified by extracting the associated modifiers and opinions. It was concluded that the recall value was lower than precision which means that certain correct feature-opinion pairs could not be recognized by the system correctly. Also, more number of features could not be extracted in negative list as most of the reviews written by users were on positive side. 3. Methodology of Proposed Work Following steps were performed in order to carry out the proposed work as follows: Figure 1: Methodology of proposed work 3.1. Select top 5 current subjects The first step is to select the top five current subjects. The subjects were selected using Google Trends [4]. Google Trends show the total search volume of a particular search term across the world. 359
- 2. International Journal of Science and Research (IJSR), India Online ISSN: 2319‐7064 Volume 2 Issue 8, August 2013 www.ijsr.net Figure 2: Selection of top five topics 3.2. Friends and Friendship Network After selecting the top five current subjects, the friends and their friends are chosen whose tweets will be retrieved in the further steps. 3.3. Develop aspect dictionary (Collect data and create bag of aspect words) This step deals with creating two separate aspect dictionaries. One contains all positive words and another contains all negative words. 3.4. Quotes, comments, opinion, extraction mining In this phase, the tweets posted by friends and friendship network on current affairs are extracted from twitter. The current affairs included are: a. Fashion b. Crime c. Safety d. Corruption e. Inflation Table 1: Formulas to find positive or negative polarity Concept Mathematical Expression Fashion Total tweets related of Fashion concept-F Positive Dictionary=Pf Negative Dictionary=Nf x words € Pf (x-no of words in Pf) y words € Nf (y-no of words in Nf) Total words belong to tweet (T1) m positive words € Pf RATIO 1= x/m=>total words in Pf/ positive words RATIO 2=R1/F= (x/m )/F Rp=Ratio for every tweet in positive dictionary=((x/m )/F)*n ∞ Positive score’Ps’= sum of all ratios= ∑((x/m )/F)*n n=1 Rn= Ratio for every tweet in negative dictionary= ((y/l)/F)*n ∞ Negative score’Ns’=sum of all ratios= ∑ ((y/l)/F)*n n=1 If(Ps>Ns)-The person has PositivePolarity/views If (Ns>Ps)-The person has Negative polarity/views Safety Total tweets related to Safety-Sa Positive Dictionary=Psa Negative Dictionary=Nsa x words € Psa (x-no of words in Psa) y words € Nsa(y-no of words in Nsa) Total words belong to tweet (T1) o positive words € Psa RATIO 1= x/o => total words in Psa/positive words RATIO 2=R1/Sa= (x/o )/Sa Rp=Ratio for every tweet in positive dictionary=((x/o )/Sa)*n ∞ Positive score’Ps’= sum of all ratio= ∑((x/o )/Sa)*n n=1 Rn= Ratio for every tweet in negative dictionary= ((y/r )/Sa)*n ∞ Negative score’Ns’= sum of all ratios= ∑ (y/r )/S)*n n=1 If(Ps>Ns)-the person has Positive Polarity/views If(Ns>Ps)-The person has Negative polarity/views Crime Total tweets related to Crime-C Positive Dictionary=Pc Negative Dictionary=Nc x words € Pc (x-no of words in Pc) y words € Nc (y-no of words in Nc) Total words belong to tweet (T1) s positive words € Pc RATIO 1= x/s => total words in Pc/positive words RATIO 2=R1/C= (x/s )/C Rp=Ratio for every tweet in positive dictionary=((x/s )/C) ∞ Positive score’Ps’= sum of all ratio= ∑((x/s )/C)*n n=1 Rn= Ratio for every tweet in negative dictionary= ((y/t)/C)*n ∞ Negative score’Ns’= sum of all ratios= ∑ ((y/t )/C)*n n=1 If(Ps>Ns)-The person has Positive Polarity/views If(Ns>Ps)-The person has Negative polarity/views Total tweets related to Fashion-CO CorruptioPositive Dictionary=Pco Negative Dictionary=Nco x words € Pco (x-no of words in Pco) y words € Nco (y-no of words in Nco) Total words belong to tweet (T1) v positive words € Pco RATIO 1= x/v => total words in Pco/positive words RATIO 2=R1/co= (x/v )/CO Rp=Ratio for every tweet in positive dictionary=((x/v )/co)*n ∞ Positive score ’Ps’= sum of all ratio= ∑((x/v )/CO)*n n=1 Rn= Ratio for every tweet in negative dictionary= ((y/w )/CO)*n ∞ Negative score ’Ns’= sum of all ratios= ∑ ((y/w )/CO)*n n=1 If(Ps>Ns)-The person has Positive Polarity/views If(Ns>Ps)-The person has Negative polarity/views Inflation Total tweets related to Fashion-I Positive Dictionary=Pi Negative Dictionary=Ni x words € Pi (x-no of words in Pi) y words € Ni (y-no of words in Ni) Total words belong to tweet (T1) q positive words € Pi RATIO 1= x/q => total words in Pi/positive words RATIO 2=R1/I= (x/q )/Pi Rp=Ratio for every tweet in positive dictionary=((x/q )/I)*n ∞ Positive score ‘Ps’= sum of all ratio= ∑((x/q )/I)*n n=1 Rn= Ratio for every tweet in negative dictionary= ((y/k)/I)*n ∞ Negative score’Ns’=sum of all ratios= ∑ ((y/k )/I*)n n=1 If(Ps>Ns)- The person has Positive Polarity/views If(Ns>Ps)-Negative polarity/views 360
- 3. International Journal of Science and Research (IJSR), India Online ISSN: 2319‐7064 Volume 2 Issue 8, August 2013 www.ijsr.net 3.5. Text parsing and storage In this phase, pre processing is done and bad words are removed from the bag of words which was created in the previous step. After performing the pre processing, the data is stored in the local text database. 3.6. Run polarity evaluation algorithm This phase deals with implementing the polarity evaluation algorithm. The input to the polarity evaluation algorithm is number and nature of positive and negative extracted, parsed, Tweets dataset and the output of this algorithm gives the polarity value in percentage. 4. Results 4.1 Fashion Results show that the positive polarity of people on fashion is 11% and negative polarity is 89%. Figure 3: Polarity values of fashion 4.2 Safety The results show that people’s views on safety have positive polarity of 86% and negative polarity of 14%. Figure 4: Polarity value of safety 4.3 Corruption The results show that positive polarity of people on corruption is 9% and negative polarity is 91%. Figure 5: Polarity values of corruption 4.4 Crime Results show that positive polarity of people on crime is 10% and negative polarity is 90%. Figure 6: Polarity values of crime 4.5 Inflation Results show that people’s views on inflation have positive polarity of 6% and negative polarity of 94%. Figure 7: Polarity values of inflation 5. Conclusion The objective of this research is to calculate the polarity of tweets posted by the celebrities. For calculating polarity tweets posted by friends and friendship network were retrieved following which bag of words were created. The polarity was calculated using an algorithm as mentioned in section. It is an objective evaluation which can be done without active participation of the people involved, and there 361
- 4. International Journal of Science and Research (IJSR), India Online ISSN: 2319‐7064 Volume 2 Issue 8, August 2013 www.ijsr.net is no breach in privacy as only public dataset of tweets is processed, and it can run continuously to bring the statistical information , what percentage of people are thinking in what way-positive or negative. It is from this views we can access the content of thoughts and their flow of attributes, feelings, mood direction of the people towards the current affairs topics like fashion, crime, safety, corruption and inflation. Future Scope In our current research we have tried to capture the views, expressions and thoughts of people using digital tools like twitter and have formulated mathematical expressions for calculating their positive aspects and negative aspects of their expressions (tweets) in an non-invasive way, which produced fairly good realistic view of how people are thinking on current aspects of things happening around them, for future scope however we suggest a deeper analysis may be conducted to extend this work. The further analysis may consist of identifying trends between various aspects of people’s opinions with identifying causal relationships between influencing parameters. This may be conducted using regression also. References [1] Nozomi Kobayashi, Kentaro Inui, and Yuji Matsumoto, “Extracting Aspect-Evaluation and Aspect-of Relations in Opinion Mining”, 2007. [2] Yi Fang, Luo Si, Naveen Somasundaram, Zhengtao Yu, “Mining Contrastive Opinions on Political Text Using Cross-Perspective Topic Model’, 2012. [3] Tanvir Ahmad, Mohammad NajmudDoja, “Ranking Sysem for Opinion Mining of Features from Review Documents”, 2012. [4] http://www.google.com/trends/ Authors Profile Manisha Punn is pursuing M. Tech (Computer Science) form Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India. She did her B. Tech (Information Technology) form Doaba Institute of Engineering and Technology, Kharar, Punjab, India. Nidhi Bhalla is working as Assistant Professor in Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India. She did her B. Tech (Information Technology) and M. Tech (Computer Science) form Lovely Professional University, Jalandhar, Punjab, India. 362