Recently, there has been an increased interest in the use of social media data as important traffic information sources and as a data research in the field of Human and Social Sciences. Social media are used, for example, to identify Twitter user communities in the context of altmetrics. In this paper, we highlight the potential use of social media data analysis by non computer science researchers. We present an approach based on a multi-dimensional analysis which combines the thematic, temporal and spatial features of tweets. We detail an experimentation using different tools in order to create the “perfectible” toolbox for researchers in HSS (for example territorial marketing) or for local territory managers. This approach can be applied in the context of altmetrics.
Twitter, a popular microblogging service, has received much attention recently. This online
social network is used by millions of people around the world to remain socially connected to
their friends, family members, and coworkers through their computers and mobile phones [
A tweet is often used as a message to friends, colleagues and other people in order to inform about situations, to share opinions, feelings, etc… about many subjects. We think we can use the information embedded in a tweet in order to identify correlations between more or less structured data.
Counting articles and citations, analyzing citations and co-authors graphs have become ways
to assess researchers and institutions performance. But, before using these ways to assess
researchers, the research data on which they are working are very important. For researchers in
the field of Human and Social Sciences, a set of tweets can be seen as a research data and
requires different steps known as the research data lifecycle: collecting, recording, processing,
analysing and then results can be published. Tweet analysis is used in many domains such as
Real-Time Event Detection, political opinion analysis, improvement of social media strategy,
marketing strategy, environment, almetrics, etc. We can quote the research work of [
We know as computer scientists that it is possible to extract, explore, synthesize and visualize knowledge based on a large mass of information available on Twitter. And sometimes, we are competent to do so. But the question is whether we can propose a framework and some tools to non computer scientist researchers to enable them to achieve these actions in a simply way. What kind of comprehensive analysis can we perform using a collection of tweets? To what end?
The goal of this paper is twofold. It first presents a framework devised from a geographical approach based on different theories and tools. Secondly it illustrates the use of dedicated tools for tweet analysis in order to identify correlations between thematic areas (what?), location of events inside the tweets or location of authors tweets (where?), time of the event (when?), and identification of sentiment.
The next section presents the general framework. 2
In this section we will present our generic framework and a toolbox we can imagine to process with this kind of research data. 2.1
The flowchart in Figure 1 depicts how tweets can be analyzed to assess the 5 W dimensions (who, when, what, where, why), three of which (When, Where and What) emphasize the geographical information. We can know who was tweeting, how, what about, what is the opinion expressed in the tweet, what kind of extraction we can do concerning a territory or a publication. As we said, for researchers in the field of Human and Social Sciences (HSS), a set of tweets can be seen as a research data. Thanks to this research data (meta data linked to the tweet or content of the tweet), researchers in the field of HSS can work on non structured textual document. For them, tweets are like others non structured textual documents: poems, epistolary exchanges between artists, ancient documents…
The main objective of the method is to analyze a collection of tweets from a multidimensional perspective. The general approach describes a process made up of different steps of analysis conducted to build a dashboard adapted to user profile: (1) preparation and validation of tweets, (2) first analysis based on user profiles and information embedded in the different fields of the tweets, (3) the multidimensional information analysis of the content of the tweet which permits the exploration of the collection of tweets and (4) summarization step 1 KPIs evaluate the success of an organization or of a particular activity (such as projects, programs, products and other initiatives) in which it engages. thanks to dashboard, map, timeline, KPI,....
After collecting the tweets we propose two types of analysis: 1. The first one is based on information mentioned in user profiles where we can analyze, in a basic way, the who, when and where dimensions, 2. The second one deals with the content of the tweet where more sophisticated analyses can be made of the 5 W dimensions.
Thanks to data in the user profiles, some statistics can be processed on the collection of tweets summarizing some of their features: number of tweets per day; users classified by location, by country; the relevant platforms, the most cited hashtags, the number of retweets,… This kind of analysis can be performed without cleaning the tweets, in a simple way. As we said, since the text is essentially informal, many challenges must be taken up in order to perform the different tweet analyses according the different dimensions. 2.2
Based on our experiments we present a toolbox that although simple, can easily be extended with new and/or improved methods. The Twitter REST search API was used for collecting tweets using GET requests. We implemented a crawler to harvest tweets automatically, every day between January 2017 and July 2017, which could contain at least the following term: #Bearn.
According to [
The classification and geocoding steps are responsible for classifying sentiment polarity and inferring geographical locations mentioned in the text, respectively.
IRAMUTEQ is an environment dedicated to lexicometry. Thanks to this environment, a
researcher can analyze much of the corpora. We can define lexicometry as the measurement of
the frequency with which words occur in text. It provides statistical indicators and graphical
representations that permit to investigate written text such as tweets. This kind of analysis
completes content analysis [
There are three levels of sentimental analysis –1) document level sentiment classification; 2)
sentence level sentiment classification; and 3) aspect level sentiment analysis. This study has
carried out a sentiment analysis by using EliXa which is an Aspect based sentiment analysis
platform developed by the Elhuyar Foundation. Given that a sentence may contain multiple
opinions, they define a window span around a given opinion target (5 words before and 5 words
after). Both scores are calculated as the sum of every positive/negative score in the
corresponding lexicon divided by the number of words in the sentence [
We have developed a processing environment with the GATE platform ([
Other processes allow us to annotate spatial entities and temporal ones. Details are given in
[
This framework and the tools described are adapted to analyse tweets in the context of Almetrics. The two types of analysis we propose on the 5W dimensions are adapted to tweets linked to set of scientific papers. Instead of waiting months, we can rapidly know the sentiment expressed in a tweet linked to a scientific paper. For example, we can measure the e-reputation of researchers thanks to positive sentiment. A major limitation is done to altmetrics findings. We do not know their meaning in the process of scholarly communication but we think almetrics can change the way to recognize the scientific contributions of researchers. .
Tweets transcript from Twitter for a period Users Profiles Tweets Assessing Who/ Where/When
Assessing «When »
Assessing Why (some
Generalities)
Assessing « About What » (Thematic Entities)
Assessing What « Sentiment analysis»
Assessing « About Where » (Spatial Entities)
First
Analysis DashBoard adapted to user profile – Summarization like maps, temporal analysis, words cloud, KPI