Spaces:
Sleeping
Attachment Bot analytics
Discuss different ways one can do analysis on the data that we collect with AttachmentBot in the Attachments and Close-Relationships lab at the University of Illinois at Urbana-Champaign.
Possible Analysis we can do:
Sentiment Analysis: You could visualize sentiment scores as histograms or density plots to see the distribution of sentiment across your chat transcripts. A time-series plot might be relevant if you're analyzing how sentiment changes over time.
Topic Modeling: For topic modeling, a common visualization is a word cloud showing the most prevalent words in each topic. Bar plots showing the distribution of topics across the dataset can also be useful.
Text Classification: If your text data is categorized into different classes, you might visualize the number of texts in each category using a bar plot or pie chart.
Named Entity Recognition (NER): You can create bar charts of the most common entities or pie charts showing the distribution of different types of entities.
Bag of Words or TF-IDF: A word cloud can be a useful visualization, showing the most common words in the dataset. A bar plot showing the words with the highest TF-IDF scores can also be insightful.
Word Embeddings: This is what we do now, Visualizing word embeddings can be challenging because they are high-dimensional, but you can use dimensionality reduction techniques like PCA or t-SNE to plot them in two or three dimensions.
n-grams Analysis: You might create bar plots of the most common n-grams or use a word cloud to visualize them.
Sequential Models like LSTM or Transformer based models: These models are complex and their outputs are not easy to visualize directly. However, you can visualize the predictions from these models or the attention scores (if applicable) to understand which parts of the text the model is focusing on for its predictions.
Potential Text Features we can look at:
Text Length: The number of words or characters in a chat message may correlate with attachment style, particularly if individuals with certain styles are more verbose or laconic.
Use of Pronouns: The use of first-person singular pronouns (e.g., "I", "me") versus first-person plural pronouns (e.g., "we", "us") could potentially indicate different attachment styles.
Sentiment Analysis: This involves determining the emotional tone behind words to gain an understanding of the attitudes, opinions, and emotions of the person communicating. People with different attachment styles may express different sentiments in their chats.
Frequency of Negative and Positive Words: Similar to sentiment analysis, you might count the number of emotionally positive and negative words used in the chat.
Lexical Diversity: This measures how many different words a person uses. It's a measure of the richness of a person's vocabulary and might be relevant to their attachment style.
Use of Attachment Words: Frequency of words specifically related to attachment, such as "close", "secure", "clingy", etc.
Topic Modeling: Topic modeling algorithms like LDA (Latent Dirichlet Allocation) could be used to identify key topics discussed in the chat. Different topics might be relevant to different attachment styles.
Question/Answer Ratio: The ratio of questions asked to statements made could be indicative of an individual's attachment style.
Attachment Figure: If your data includes information about the attachment figure being discussed (e.g., mother, father, partner), this could be a very useful feature. Different people may have different attachment styles with different figures.