Download Sentiment Analysis on Twitter and more Thesis Computer Science in PDF only on Docsity!
Sentiment analysis on twitter
Report submitted for the partial fulfillment of the requirements for the degree of
Bachelor of Technology in
Information Technology
Submitted by
Name and Roll Number
Avijit Pal (IT2014/052)
Argha Ghosh (IT2014/056) 11700214016
Bivuti Kumar (IT2014/061) 11700214083
Under the Guidance of Mr. Amit Khan (Assistant Professor(IT),RCCIIT)
RCC Institute of Information Technology
Canal South Road, Beliaghata, Kolkata – 700 015
[Affiliated to West Bengal University of Technology]
Acknowledgement
We would like to express our sincere gratitude to Mr. Amit khan of the department of Information Technology, whose role as project guide was invaluable for the project. We are extremely thankful for the keen interest he took in advising us, for the books and reference materials provided for the moral support extended to us.
Last but not the least we convey our gratitude to all the teachers for providing us the technical skill that will always remain as our asset and to all non-teaching staff for the gracious hospitality they offered us.
Place: RCCIIT, Kolkata
Date: 14.05.
Avijit Pal
Argha Ghosh
Bivuti Kumar
Department of Information Technology RCCIIT, Beliaghata, Kolkata – 700 015, West Bengal, India
INDEX:
Contents
**_1. Introduction
- Problem Definition
- Literature Survey
- SRS (Software Requirement Specification)
- Design_**
6.Results and Discussion
7. Conclusion and Future Scope
6.Reference
Page Numbers
5
7
8
14
Introduction:
In the past few years, there has been a huge growth in the use of microblogging platforms such as Twitter. Spurred by that growth, companies and media organizations are increasingly seeking ways to mine Twitter for information about what people think and feel about their products and services. Companies such as Twitratr (twitrratr.com), tweetfeel (www.tweetfeel.com), and Social Mention (www.socialmention.com) are just a few who advertise Twitter sentiment analysis as one of their services.
While there has been a fair amount of research on how sentiments are expressed in genres such as online reviews and news articles, how sentiments are expressed given the informal language and message-length constraints of microblogging has been much less studied. Features such as automatic part-of-speech tags and resources such as sentiment lexicons have proved useful for sentiment analysis in other domains, but will they also prove useful for sentiment analysis in Twitter? In this paper, we begin to investigate this question.
Another challenge of microblogging is the incredible breadth of topic that is covered. It is not an exaggeration to say that people tweet about anything and everything. Therefore, to be able to build systems to mine Twitter sentiment about any given topic, we need a method for quickly identifying data that can be used for training. In this paper, we explore one method for building such data: using Twitter hashtags (e.g., #bestfeeling, #epicfail, #news) to identify positive, negative, and neutral tweets to use for training threeway sentiment classifiers.
The online medium has become a significant way for people to express their opinions and with social media, there is an abundance of opinion information available. Using sentiment analysis, the polarity of opinions can be found, such as positive, negative, or neutral by analyzing the text of the opinion. Sentiment analysis has been useful for companies to get their customer's opinions on their products predicting outcomes of elections , and getting opinions from movie reviews. The information gained from sentiment analysis is useful for companies making future decisions. Many traditional approaches in sentiment analysis uses the bag of words method. The bag of words technique does not consider language morphology, and it could incorrectly classify two phrases of having the same meaning because it could have the same bag of words. The relationship between the collection of words is considered instead of the relationship between individual words. When determining the overall sentiment, the sentiment of each word is determined and combined using a function. Bag of words also ignores word order, which leads to phrases with negation in them to be incorrectly classified. Other techniques discussed in sentiment analysis include Naive Bayes, Maximum Entropy, and Support Vector Machines. In the Literature Survey section, approaches used for sentiment analysis and text classification are summarized.
Sentiment analysis refers to the broad area of natural language processing which deals with the computational study of opinions, sentiments and emotions expressed in text. Sentiment Analysis (SA) or Opinion Mining (OM) aims at learning people’s opinions, attitudes and emotions towards an entity. The entity can represent individuals, events or topics. An immense amount of research has been performed in the area of sentiment analysis. But most of them focused on classifying formal and larger pieces of text data like reviews. With the wide popularity of social networking
Problem Definition:
Sentiment analysis of in the domain of micro-blogging is a relatively new research topic so there is still a lot of room for further research in this area. Decent amount of related prior work has been done on sentiment analysis of user reviews , documents, web blogs/articles and general phrase level sentiment analysis. These differ from twitter mainly because of the limit of 140 characters per tweet which forces the user to express opinion compressed in very short text. The best results reached in sentiment classification use supervised learning techniques such as Naive Bayes and Support Vector Machines, but the manual labelling required for the supervised approach is very expensive. Some work has been done on unsupervised and semi-supervised approaches, and there is a lot of room of improvement. Various researchers testing new features and classification techniques often just compare their results to base-line performance. There is a need of proper and formal comparisons between these results arrived through different features and classification techniques in order to select the best features and most efficient classification techniques for particular applications.
Literature Survey:
Sentiment analysis is a growing area of Natural Language Processing with research ranging from document level classification (Pang and Lee 2008) to learning the polarity of words and phrases (e.g., (Hatzivassiloglou and McKeown 1997; Esuli and Sebastiani 2006)). Given the character limitations on tweets, classifying the sentiment of Twitter messages is most similar to sentence- level sentiment analysis (e.g., (Yu and Hatzivassiloglou 2003; Kim and Hovy 2004)); however, the informal and specialized language used in tweets, as well as the very nature of the microblogging domain make Twitter sentiment analysis a very different task. It’s an open question how well the features and techniques used on more well-formed data will transfer to the microblogging domain.
Just in the past year there have been a number of papers looking at Twitter sentiment and buzz (Jansen et al. 2009 ; Pak and Paroubek 2010; O’Connor et al. 2010; Tumasjan et al. 2010; Bifet and Frank 2010; Barbosa and Feng 2010 ; Davidov, Tsur, and Rappoport 2010). Other researchers have begun to explore the use of part-of-speech features but results remain mixed. Features common to microblogging (e.g., emoticons) are also common, but there has been little investigation into the usefulness of existing sentiment resources developed on non-microblogging data.
Researchers have also begun to investigate various ways of automatically collecting training data. Several researchers rely on emoticons for defining their training data (Pak and Paroubek 2010; Bifet and Frank 2010). (Barbosa and Feng 2010) exploit existing Twitter sentiment sites for collecting training data. (Davidov, Tsur, and Rappoport 2010) also use hashtags for creating training data, but they limit their experiments to sentiment/non-sentiment classification, rather than 3-way polarity classification, as we do. We use WEKA and apply the following Machine Learning algorithms for this second classification to arrive at the best result:
- K-Means Clustering
- Support Vector Machine
- Logistic Regression
- K Nearest Neighbours
- Naive Bayes
- Rule Based Classifiers
The algoritham of sentiment analysis using Naive Bayes Classification:
function BOOTSTRAP(x,b) returns p-value(x)
Calculate δ(x)
for i = 1to b do
for j = 1to n do # Draw a bootstrap sample x∗(i) of size n
Select a member of x at random and add it to x∗(i)
Calculate δ(x∗(i))
For each x∗(i)
s←s + 1ifδ(x∗(i)) > 2δ(x)
p-value(x)≈ s b
return p-value(x)
- Many language processing tasks can be viewed as tasks of classification. learn to model the class given the observation.
- Textcategorization,inwhichanentiretextisassignedaclassfromafiniteset, comprises such tasks as sentiment analysis, spam detection, email classification, and authorship attribution.
- Sentimentanalysisclassifiesatextasreflectingthepositiveornegativeorientation (sentiment) that a writer expresses toward some object.
- Naive Bayes is a generative model that make the bag of words assumption (positiondoesn’tmatter)andtheconditionalindependenceassumption(words are conditionally independent of each other given the class)
- Naive Bayes with binarized features seems to work better for many text classification tasks.
The TextBlob package for Python is a convenient way to do a lot of Natural Language Processing (NLP) tasks. For example:
From textblob import TextBlob
TextBlob(“not a very great calculation”).sentiment
This tells us that the English phrase “not a very great calculation” has a polarity of about -0.3, meaning it is slightly negative, and a subjectivity of about 0.6, meaning it is fairly subjective.
There are helpful comments like this one, which gives us more information about the numbers we're interested in:
Each word in the lexicon has scores for:
1) polarity: negative vs. positive (-1.0 => +1.0)
2) subjectivity: objective vs. subjective (+0.0 => +1.0)
3) intensity: modifies next word? (x0.5 => x2.0)
The lexicon it refers to is in en-sentiment.xml, an XML document that includes the following four entries for the word “great”.
<word Form=”great” cornetto svnset id=”n_a-525317” wordnet id=”a-01123879” pos=”JJ” sense=”very good” polanty=”1.0” subjectivity=”1.0” intensity=”1.0” confidence=”0.9” />
<word Form=”great” wordnet id=”a-011238818” pos=”JJ” sense=”of major significance or importance” polanty=”1.0” subjectivity=”1.0” intensity=”1.0” confidence=”0.9” />
<word Form=”great” wordnet id=”a-01123883” pos=”JJ” sense=”relativity large in size or number or extent” polanty=”0.4” subjectivity=”0.2” intensity=”1.0” confidence=”0.9” />
<word Form=”great” wordnet id=”a-01677433” pos=”JJ” sense=”remarkable or out of the ordinary in degree or magnitude or effect” polanty=”0.8” subjectivity=”0.8” intensity=”1.0” confidence=”0.9” />
In addition to the polarity, subjectivity, and intensity mentioned in the comment above, there's also “confidence”, but I don't see this being used anywhere. In the case of “great” here it's all the same part of speech (JJ, adjective), and the senses are themselves natural language and not used. To simplify for readability:
Word Polarity Subjectivity Intensity
Great 1.0 1.0 10
Great 1.0 1.0 1.
Great 0.4 0.2 1.
Great 0.8 0.8 1.
When calculating sentiment for a single word, TextBlob uses a sophisticated technique known to
Mathematicians as “averaging”.
TextBlob(“great").sentiment
Sentiment(polarity=0.8, subjectivity=0.75)
At this point we might feel as if we're touring a sausage factory. That feeling isn't going to go away, but remember how delicious sausage is! Even if there isn't a lot of magic here, the results can be useful—and you certainly can't beat it for convenience.
TextBlob doesn't not handle negation, and that ain't nothing!
And while I'm being a little critical, and such a system of coded rules is in some ways the antithesis of machine learning, it is still a pretty neat system and I think I'd be hard-pressed to code up a better such solution.
SRS (Software Requirement Specification):
Internal Interface requirement: Identify the product whose software requirements are specified in this document, including the revision or release number. Describe the scope of the product that is covered by this SRS, particularly if this SRS describes only part of the system or a single subsystem. Describe any standards or typographical conventions that were followed when writing this SRS, such as fonts or highlighting that have special significance. For example, state whether priorities for higher-level requirements are assumed to be inherited by detailed requirements, or whether every requirement statement is to have its own priority. Describe the different types of reader that the document is intended for, such as developers, project managers, marketing staff, users, testers, and documentation writers. Describe what the rest of this SRS contains and how it is organized. Suggest a sequence for reading the document, beginning with the overview sections and proceeding through the sections that are most pertinent to each reader type. Provide a short description of the software being specified and its purpose, including relevant benefits, objectives, and goals. Relate the software to corporate goals or business strategies. If a separate vision and scope document is available, refer to it rather than duplicating its contents here. The recent explosion in data pertaining to users on social media has created a great interest in performing sentiment analysis on this data using Big Data and Machine Learning principles to understand people's interests. This project intends to perform the same tasks. The difference between this project and other sentimnt analysis tools is that, it will perform real time analysis of tweets based on hashtags and not on a stored archive.
;Describe the context and origin of the product being specified in this SRS. For example, state whether this product is a ;follow-on member of a product family, a replacement for certain existing systems, or a new, self-contained product. If the ;SRS defines a component of a larger system, relate the requirements of the larger system to the functionality of this ;software and identify interfaces between the two. A simple diagram that shows the major components of the overall system, ;subsystem interconnections, and external interfaces can be helpful.
The Product functions are:
- Collect tweets in a real time fashion i.e. , from the twitter live stream based on specified hashtags
- Remove redundant information from these collected tweets.
- Store the formatted tweets in MongoDB database
- Perform Sentiment Analysis on the tweets stored in the database to classify their nature viz. positive, negative and so on.
- Use a machine learning algorithm which will predict the ‘mood’ of the people with respect ot that topic.
;Summarize the major functions the product must perform or must let the user perform. Details will be provided in Section 3, ;so only a high level summary (such as a bullet list) is needed here. Organize the functions to make them understandable to ;any reader of the SRS. A picture of the
example, use of a global data area in a multitasking operating system), specify this as an
implementation constraint.
Communication Interface:
Describe the requirements associated with any communications functions required by this product, including e-mail, web browser, network server communications protocols, electronic forms, and so on. Define any pertinent message formatting. Identify any communication standards that will be used, such as FTP or HTTP. Specify any communication security or encryption issues, data transfer rates, and synchronization mechanisms.
Non Functional Requirement:
Performance Requirements:
If there are performance requirements for the product under various circumstances, state them here and explain their rationale, to help the developers understand the intent and make suitable design choices. Specify the timing relationships for real time systems. Make such requirements as specific as possible. You may need to state performance requirements for individual functional requirements or features.
Safety Requirements:
Specify those requirements that are concerned with possible loss, damage, or harm that could result from the use of the product. Define any safeguards or actions that must be taken, as well as actions that must be prevented. Refer to any external policies or regulations that state safety issues that affect the product’s design or use. Define any safety certifications that must be satisfied.
Security Requirements:
Specify any requirements regarding security or privacy issues surrounding use of the product or protection of the data used or created by the product. Define any user identity authentication requirements. Refer to any external policies or regulations containing security issues that affect the product. Define any security or privacy certifications that must be satisfied.
Software Quality Attributes:
Specify any additional quality characteristics for the product that will be important to either the customers or the developers. Some to consider are: adaptability, availability, correctness, flexibility, interoperability, maintainability, portability, reliability, reusability, robustness, testability, and usability. Write these to be specific, quantitative, and verifiable when possible. At
the least, clarify the relative preferences for various attributes, such as ease of use over ease of learning.
Other Requirements:
Linux Operating System/Windows Python Platform(Anaconda2,Spyder,Jupyter) NLTK package, Modern Web Browser Twitter API, Google API
EYWORD)
Input(KeyWord):
Data in the form of raw tweets is acquired by using the Python library “tweepy” which provides a package for simple twitter streaming API. This API allows two modes of accessing tweets: SampleStream and FilterStream. SampleStream simply delivers a small, random sample of all the tweets streaming at a real time. FilterStream delivers tweet which match a certain criteria. It can filter the delivered tweets according to three criteria:
- Specific keyword to track/search for in the tweets
- Specific Twitter user according to their name
- Tweets originating from specific location(s) (only for geo-tagged tweets). A programmer can specify any single one of these filtering criteria or a multiple combination of these. But for our purpose we have no such restriction and will thus stick to the SampleStream mode. Since we wanted to increase the generality of our data, we acquired it in portions at different points of time instead of acquiring all of it at one go. If we used the latter approach then the generality of the tweets might have been compromised since a significant portion of the tweets would be referring to some certain trending topic and would thus have more or less of the same general mood or sentiment. This phenomenon has been observed when we were going through our sample of acquired tweets. For example the sample acquired near Christmas and New Year’s had a significant portion of tweets referring to these joyous events and were thus of a generally positive sentiment. Sampling our data in portions at different points in time would thus try to minimize this problem. Thus forth, we acquired data at four different points which would be 17th of December 2015, 29th of December 2015, 19th of January 2016 and 8th of February 2016. A tweet acquired by this method has a lot of raw information in it which we may or may not find useful for our particular application. It comes in the form of the python “dictionary” data type with various key-value pairs. A list of some key-value pairs are given below:
- Whether a tweet has been favourited
- User ID
- Screen name of the user
- Original Text of the tweet
- Presence of hashtags
- Whether it is a re-tweet
- Language under which the twitter user has registered their account
- Geo-tag location of the tweet
- Date and time when the tweet was created
Since this is a lot of information we only filter out the information that we need and discard the rest. For our particular application we iterate through all the tweets in our sample and save the actual text content of the tweets in a separate file given that language of the twitter is user’s account
is specified to be English. The original text content of the tweet is given under the dictionary key “text” and the language of user’s account is given under “lang”.
TWEETS RETRIEVAL Tweets Retrieval:
Since human labelling is an expensive process we further filter out the tweets to be labelled so that we have the greatest amount of variation in tweets without the loss of generality. The filtering criteria applied are stated below:
- Remove Retweets (any tweet which contains the string “RT”)
- Remove very short tweets (tweet with length less than 20 characters)
- Remove non-English tweets (by comparing the words of the tweets with a list of 2,000 common English words, tweets with less than 15% of content matching threshold are discarded)
- Remove similar tweets (by comparing every tweet with every other tweet, tweets with more than 90% of content matching with some other tweet is discarded)
After this filtering roughly 30% of tweets remain for human labelling on average per sample, which made a total of 10,173 tweets to be labelled.
Data Preprocessing:
Data preprocessing consists of three steps:
- tokenization,
- normalization, and
- part-of-speech (POS) tagging.
Tokenization:
It is the process of breaking a stream of text up into words, symbols and other meaningful elements called “tokens”. Tokens can be separated by whitespace characters and/or punctuation characters. It is done so that we can look at tokens as individual components that make up a tweet.Emoticons and abbreviations (e.g., OMG, WTF, BRB ) are identified as part of the tokenization process and treated as individual tokens.
Normalization:
For the normalization process, the presence of abbreviations within a tweet is noted and then abbreviations are replaced by their actual meaning (e.g., BRB − > be right back ). We also identify informal intensifiers such as all-caps (e.g., I LOVE this show!!! and character repetitions (e.g., I’ve got a mortgage!! happyyyyyy” ), note their presence in the tweet. All-caps words are made into lower case, and instances of repeated characters are replaced by a single character. Finally, the
