Scribd
Upload
121 views2 pages

Hands On Text Analytics With Orange - Digital Humanities 2017

This document outlines the structure and content of a workshop on using the Orange data mining environment for text analytics. The workshop is divided into three parts: 1) data import and preprocessing techniques demonstrated on sample corpora, 2) predictive modeling using machine learning classifiers on Twitter data after preprocessing, and 3) additional techniques like clustering, sentiment analysis, and image/geo analytics on social media data. Throughout, participants will construct analytical workflows and apply techniques like embedding, visualization, and machine learning to expose patterns in text and image data.

Uploaded by

Yas@786
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
121 views2 pages

Hands On Text Analytics With Orange - Digital Humanities 2017

This document outlines the structure and content of a workshop on using the Orange data mining environment for text analytics. The workshop is divided into three parts: 1) data import and preprocessing techniques demonstrated on sample corpora, 2) predictive modeling using machine learning classifiers on Twitter data after preprocessing, and 3) additional techniques like clustering, sentiment analysis, and image/geo analytics on social media data. Throughout, participants will construct analytical workflows and apply techniques like embedding, visualization, and machine learning to expose patterns in text and image data.

Uploaded by

Yas@786
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

be

introduced to several options for data import, from


standard Corpus to Twitter, Guardian and Text Import.
Once the corpus is loaded, we will preprocess it and
Hands on Text Analytics display the result in a word cloud. A particular empha-
sis will be on the use of custom preprocessing tech-
with Orange niques and how to successfully apply them to the cor-
pus. The results of each technique will be observed in
Ajda Pretnar an interactive word cloud and concordances.
[email protected]
University of Ljubljana, Slovenia

Niko Colnerič
[email protected]
University of Ljubljana, Slovenia

Lan Žagar
[email protected]
University of Ljubljana, Slovenia


Orange for Text Analytics
Figure 1: Preprocessing results displayed in a word cloud
In recent years, the digital humanities community
Part 2: Machine learning and deep-learning-
has been introduced to many powerful tools for text
analysis, but few of these tools combine powerful data based embedding for predictive analysis
mining and machine learning algorithms within a sim- Next, we will use Twitter data to construct an au-
ple and capable user interface. For flexible and crea- thor prediction pipeline and test some classifiers. We
tive analysis, researchers need a tool that focuses on will fetch author Timelines from Twitter and observe
intuition, visualizations and interactivity. the retrieved corpus. This time we will introduce a
This workshop will introduce participants to Or- pre-trained tweet tokenizer and pass the prepro-
ange, a visual programming environment for data min- cessed corpus through a bag of words. We will discuss
ing, suitable for both beginners and experts. Particular bag of words parameters and how to best prepare the
emphasis will be placed on its Text add-on, which of- data for further analysis. The results of using different
fers components for text mining, visualization and parameters will be observed in a data table to under-
deep- learning-based embedding. stand the underlying data structures. For comparison,
This is a hands-on workshop, where the partici- we will use deep-learning-based embedding to derive
pants will actively construct analytical workflows and vector representation of tweets and in this way enable
go through case studies with the help of the instruc- machine learning.
tors. They will learn how to manage textual data, pre- We will explain how we can use machine learning
process it, use machine learning, data projection and in text mining and introduce a number of techniques
visualisation techniques to expose hidden patterns for predictive analysis. We will use cross-validation to
and evaluate the resulting models. At the end of the test the constructed bag of words models and compare
workshop, the participants will know how to use vis- classification scores for each algorithm. We will dis-
ual programming to seamlessly construct powerful cuss the quality of constructed models and what
data analysis workflows, which can be applied to a scores are usually the best for observing model qual-
wide range of challenges in digital humanities. ity. Additionally, we will inspect misclassified tweets
in a confusion matrix and even further in Corpus
Structure of the Workshop
Viewer, to leverage the possibilities of a close(r) read-
Part 1: Visual programming, workflows, data ing.
input and preprocessing
Part 3: Data clustering, sentiment analysis,
First, we will show the basics of Orange: how to image and geo analytics
load the data, inspect and visualize it. Participants will
In the third part, we will work on geomapping and
image analytics. We will transform textual and visual
data into feature vectors and plot these data onto a
world map to discover interesting relations.
We will discuss how to acquire geolocated data
from Twitter and why this is useful. Next, we will use
geotagged Twitter data and apply a pre-trained senti-
ment analysis model to acquire sentiment orientation.
We will map the sentiment-tagged tweets and explore
how to use sentiment together with geomapping.
Finally, the participants will be introduced to image
analytics for humanities research. We will explain why
and how to transform raw images into multidimen-
sional vectors and how to work with the new data. We
will cluster Instagram images into groups and explore
how to map image-containing tweets on a world map.
Do images correspond to geolocation? We will see.


Figure 2: Images from social media are embedded with
ImageNet embedding, clustered with Hierarchical Clustering
and displayed on a map by their geolocation.

You might also like