Skip to content

Latest commit

 

History

History
72 lines (50 loc) · 2.8 KB

README.md

File metadata and controls

72 lines (50 loc) · 2.8 KB
Raw

Blickwinkel R Package

Batch Sentiment Analysis for German

License Version

Overview

This project implements a batch sentiment analysis tool for the German language, leveraging the Leipzig SentiWS dataset to assign sentiment scores to sentences.

Installation

You can install the blickwinkel package directly from GitHub using the devtools package:

# Install devtools if you haven't already
install.packages("devtools")

# Install blickwinkel from GitHub
devtools::install_github("DanielGall500/blickwinkel")

Data Sources

Leipzig SentiWS Dataset The Leipzig SentiWS dataset provides sentiment scores for German lexical items. Features include:

  • Sentiment Score Range: -1 (very negative) to +1 (very positive).
  • Part-of-Speech (POS) Tags for each term.
  • Alternate Forms of lexical items for comprehensive coverage.

Research Objective

The aim is to determine which named entities appear in the most positive and negative contexts. By analyzing a subset of 260 sentences, we can:

  1. Test the effectiveness of the sentiment analysis tool.
  2. Provide insights into sentiment distributions around named entities in German text.

Methodology

1. Preprocessing

  • Lexical Item Separation: The SentiWS dataset is split into lexical items and their POS tags.
  • Sentence Data Loading: The treebank data is formatted to identify named entities and their surrounding contexts.
  • Document-Term Matrix (DTM): Created using the tm library in R to map terms to their respective sentences.

2. Matrix-Based Sentiment Analysis

The analysis uses matrix operations to batch-process sentiment scores:

  1. Combine Forms: Group lexical items with their alternate forms, consolidating their sentiment values into a single column in the matrix.
  2. Apply Sentiment Scores: Multiply the sentiment scores of terms by their occurrence frequency in sentences using column-wise matrix operations.
  3. Generate Sentiment Matrix: A sparse matrix is created, where:
    • Rows represent sentences.
    • Columns represent sentiment-relevant terms.
    • Values correspond to sentiment scores for each term-sentence pair.

3. Final Sentiment Calculation

  • Row sums of the sentiment matrix provide overall sentiment scores for each sentence.
  • The dataset is then enriched with these scores for further analysis.

Applications

This methodology can be applied to:

  1. Social Media Analysis: Detect sentiment trends towards entities like brands, public figures, or events.
  2. Market Research: Gauge public opinion from large text corpora.
  3. Linguistic Research: Understand sentiment dynamics in different languages and contexts.