Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/emilyfelker/bloganalysis

Can ChatGPT estimate my age? Program to extract and analyze personal blog posts with basic features and ChatGPT
https://github.com/emilyfelker/bloganalysis

chatgpt chatgpt-api generative-ai natural-language-processing nlp poetry pytest python sql sqlite3

Last synced: 8 days ago
JSON representation

Can ChatGPT estimate my age? Program to extract and analyze personal blog posts with basic features and ChatGPT

Awesome Lists containing this project

README 

        
# Blog Analysis with ChatGPT



## Introduction



This program analyzes a set of around 800 blog posts, mostly about my 

personal life, that I wrote during my teenage years and into college. It uses

`BeautifulSoup` to parse the HTML files that constitute the blog archive

and extract each post's title, date, and body. Various features of interest 

about each blog post are calculated. For some features, ChatGPT is called 

via the `OpenAI` API, and its responses are cached in a `sqlite3` database

to avoid unnecessary API calls when re-running the program.

The program uses `Matplotlib` and `NumPy` to create graphs that show how

certain numerical features change over time. One question I had was: based

on each blog post's text, how 

accurately can ChatGPT estimate my age?



Additionally, the project implements 

**Retrieval-Augmented Generation 

(RAG)** with `LangChain` to 

allow for natural language queries 

about the blog posts, which are 

split into smaller chunks for better 

retrieval and embedded with 

`OpenAIEmbeddings`. This 

functionality uses `FAISS` 

for vector-based similarity search and leverages OpenAI's GPT model to generate 

answers based on the retrieved blog post content.



## RAG Usage Examples



Example interaction with the 

interactive question-answering tool:

```

Welcome to the Blog Analysis Query Tool!

Enter your query (or type 'exit' to quit): What were my career aspirations back then?

Answer:

Your career aspirations included becoming a librarian or library technician, 

as well as exploring service-related jobs and computer-related careers. You 

also expressed an interest in becoming a research psychologist, although you 

were uncertain about your future direction. Ultimately, you planned to major 

in psychology in college, with the possibility of switching to other interests 

such as French language, library science, computer science, or political science. 

You aimed to pursue graduate and doctorate degrees and sought a job associated 

with a university or a federal government position.



Enter your query (or type 'exit' to quit): exit

Goodbye!

```



Answers can also be generated 

programatically by using the RAG 

functions directly in Python:

 ```python

from BlogAnalysis.rag import generate_answer



question = "Who were Emily's lovers?"

answer = generate_answer(question)

print(f"Answer:\n{answer}")

```



## Data Visualization



This scatterplot of Chat GPT's age estimate plotted against my actual age shows (thankfully!) a positive correlation between the two variables, with a few outliers:

![Scatterplot with linear regression line showing a positive correlation between age estimate and actual age](output/real_data_graphed.png "Actual Age vs. GPT Estimate")



In the future, the program could be expanded to see whether my age at the time of writing correlates better with more traditional predictors like word count or difficulty, sentence length, or other measures of writing complexity.



## Data Analysis Examples



Loading the dataset and calculating features:

```python

dataset = get_blogpost_dataset("data/XangaBlogPosts")

dataset_with_features = add_features_to_dataset(dataset)

```

Previewing the post title, beginning of post body, and features:

```python

preview_features(dataset_with_features[:10]  # just the first ten posts

```

Which prints output like this per post:

```

Title: Which Language to Learn Next? | Date: 2012-02-07

Content: After seven years of study, my level of French seems to have reached a point of ...

Features:

  word_count: 743

  day_of_week: Tuesday

  age_of_emily: 21.078713210130047

  topic: Choosing between German and Spanish

  age_estimate: 30.0

```

For a fun trip down memory lane, I also wanted a quick and easy way to look at just the topic

summaries ChatGPT generated for all my posts:

```python

show_summaries(dataset_with_features[:10])  # just the first ten posts

```

Which prints output like this:

```

Choosing between German and Spanish

First Christmas away from family.

Highlights of trip: living with cats

Visiting Pretoria and its attractions.

Professor uses powerful songs, poems.

Parents and others demand smiles

Pet peeves on self-referential posts.

Regret ending online friendship, loneliness

Challenging senior year with rigorous courses

Anxiety-induced sleep troubles and volunteering.

Unhappy with P.E. teacher, election excitement, terrified of squirrels

Dream of exploring Chinese city.

Physics project success, English class discomfort, unexpected reunion

Key Club election results and socializing

WYSE competition and school fundraiser.

Photos from colorful park outing.

```