Tutorial: Semantic Clustering of Consumer Messages with LLM Prompts

As a Developer Advocate, it’s difficult to maintain up with consumer discussion board messages and perceive the large image of what customers are saying. There’s loads of helpful content material — however how are you going to shortly spot the important thing conversations? On this tutorial, I’ll present you an AI hack to carry out semantic clustering just by prompting LLMs!

TL;DR 🔄 this weblog submit is about the way to go from (information science + code) → (AI prompts + LLMs) for a similar outcomes — simply sooner and with much less effort! 🤖⚡. It’s organized as follows:

• Inspiration and Knowledge Sources
• Exploring the Knowledge with Dashboards
• LLM Prompting to provide KNN Clusters
• Experimenting with Customized Embeddings
• Clustering Throughout A number of Discord Servers

Inspiration and Knowledge Sources

First, I’ll give props to the December 2024 paper Clio (Claude insights and observations), a privacy-preserving platform that makes use of AI assistants to research and floor aggregated utilization patterns throughout tens of millions of conversations. Studying this paper impressed me to do that.

Knowledge. I used solely publicly obtainable Discord messages, particularly “discussion board threads”, the place customers ask for tech assist. As well as, I aggregated and anonymized content material for this weblog.  Per thread, I formatted the information into dialog flip format, with consumer roles recognized as both “consumer”, asking the query or “assistant”, anybody answering the consumer’s preliminary query. I additionally added a easy, hard-coded binary sentiment rating (0 for “not joyful” and 1 for “joyful”) primarily based on whether or not the consumer mentioned thanks anytime of their thread. For vectorDB distributors I used Zilliz/Milvus, Chroma, and Qdrant.

Step one was to transform the information right into a pandas information body. Beneath is an excerpt. You possibly can see for thread_id=2, a consumer solely requested 1 query. However for thread_id=3, a consumer requested 4 totally different questions in the identical thread (different 2 questions at farther down timestamps, not proven under).

I added a naive sentiment 0|1 scoring perform.

def calc_score(df):
   # Outline the goal phrases
   target_words = ["thanks", "thank you", "thx", "🙂", "😉", "👍"]


   # Helper perform to verify if any goal phrase is within the concatenated message content material
   def contains_target_words(messages):
       concatenated_content = " ".be part of(messages).decrease()
       return any(phrase in concatenated_content for phrase in target_words)


   # Group by 'thread_id' and calculate rating for every group
   thread_scores = (
       df[df['role_name'] == 'consumer']
       .groupby('thread_id')['message_content']
       .apply(lambda messages: int(contains_target_words(messages)))
   )
   # Map the calculated scores again to the unique DataFrame
   df['score'] = df['thread_id'].map(thread_scores)
   return df


...


if __name__ == "__main__":
  
   # Load parameters from YAML file
   config_path = "config.yaml"
   params = load_params(config_path)
   input_data_folder = params['input_data_folder']
   processed_data_dir = params['processed_data_dir']
   threads_data_file = os.path.be part of(processed_data_dir, "thread_summary.csv")
  
   # Learn information from Discord Discussion board JSON information right into a pandas df.
   clean_data_df = process_json_files(
       input_data_folder,
       processed_data_dir)
  
   # Calculate rating primarily based on particular phrases in message content material
   clean_data_df = calc_score(clean_data_df)


   # Generate reviews and plots
   plot_all_metrics(processed_data_dir)


   # Concat thread messages & save as CSV for prompting.
   thread_summary_df, avg_message_len, avg_message_len_user = 
   concat_thread_messages_df(clean_data_df, threads_data_file)
   assert thread_summary_df.form[0] == clean_data_df.thread_id.nunique()

Exploring the Knowledge with Dashboards

From the processed information above, I constructed conventional dashboards:

• Message Volumes: One-off peaks in distributors like Qdrant and Milvus (probably resulting from advertising occasions).
• Consumer Engagement: High customers bar charts and scatterplots of response time vs. variety of consumer turns present that, typically, extra consumer turns imply increased satisfaction. However, satisfaction does NOT look correlated with response time. Scatterplot darkish dots appear random with regard to y-axis (response time). Possibly customers should not in manufacturing, their questions should not very pressing? Outliers exist, comparable to Qdrant and Chroma, which can have bot-driven anomalies.
• Satisfaction Tendencies: Round 70% of customers seem joyful to have any interplay. Knowledge notice: be sure to verify emojis per vendor, typically customers reply utilizing emojis as a substitute of phrases! Instance Qdrant and Chroma.

LLM Prompting to provide KNN Clusters

For prompting, the following step was to mixture information by thread_id. For LLMs, you want the texts concatenated collectively. I separate out consumer messages from total thread messages, to see if one or the opposite would produce higher clusters. I ended up utilizing simply consumer messages.

With a CSV file for prompting, you’re able to get began utilizing a LLM to do information science!

!pip set up -q google.generativeai
import os
import google.generativeai as genai


# Get API key from native system
api_key=os.environ.get("GOOGLE_API_KEY")


# Configure API key
genai.configure(api_key=api_key)


# Checklist all of the mannequin names
for m in genai.list_models():
   if 'generateContent' in m.supported_generation_methods:
       print(m.title)


# Attempt totally different fashions and prompts
GEMINI_MODEL_FOR_SUMMARIES = "gemini-2.0-pro-exp-02-05"
mannequin = genai.GenerativeModel(GEMINI_MODEL_FOR_SUMMARIES)
# Mix the immediate and CSV information.
full_input = immediate + "nnCSV Knowledge:n" + csv_data
# Inference name to Gemini LLM
response = mannequin.generate_content(full_input)


# Save response.textual content as .json file...


# Verify token counts and examine to mannequin restrict: 2 million tokens
print(response.usage_metadata)

Sadly Gemini API stored chopping quick the response.textual content. I had higher luck utilizing AI Studio instantly.

My 5 prompts to Gemini Flash & Professional (temperature set to 0) are under.

Immediate#1: Get thread Summaries:

Given this .csv file, per row, add 3 columns:
– thread_summary = 205 characters or much less abstract of the row’s column ‘message_content’
– user_thread_summary = 126 characters or much less abstract of the row’s column ‘message_content_user’
– thread_topic = 3–5 phrase tremendous high-level class
Ensure the summaries seize the principle content material with out shedding an excessive amount of element. Make consumer thread summaries straight to the purpose, seize the principle content material with out shedding an excessive amount of element, skip the intro textual content. If a shorter abstract is sweet sufficient desire the shorter abstract. Ensure the subject is normal sufficient that there are fewer than 20 high-level subjects for all the information. Choose fewer subjects. Output JSON columns: thread_id, thread_summary, user_thread_summary, thread_topic.

Immediate#2: Get cluster stats:

Given this CSV file of messages, use column=’user_thread_summary’ to carry out semantic clustering of all of the rows. Use approach = Silhouette, with linkage methodology = ward, and distance_metric = Cosine Similarity. Simply give me the stats for the strategy Silhouette evaluation for now.

Immediate#3: Carry out preliminary clustering:

Given this CSV file of messages, use column=’user_thread_summary’ to carry out semantic clustering of all of the rows into N=6 clusters utilizing the Silhouette methodology. Use column=”thread_topic” to summarize every cluster subject in 1–3 phrases. Output JSON with columns: thread_id, level0_cluster_id, level0_cluster_topic.

Silhouette Rating measures how comparable an object is to its personal cluster (cohesion) versus different clusters (separation). Scores vary from -1 to 1. A better common silhouette rating usually signifies better-defined clusters with good separation. For extra particulars, try the scikit-learn silhouette rating documentation.

Making use of it to Chroma Knowledge. Beneath, I present outcomes from Immediate#2, as a plot of silhouette scores. I selected N=6 clusters as a compromise between excessive rating and fewer clusters. Most LLMs nowadays for information evaluation take enter as CSV and output JSON.

From the plot above, you possibly can see we’re lastly moving into the meat of what customers are saying!

Immediate#4: Get hierarchical cluster stats:

Given this CSV file of messages, use the column=’thread_summary_user’ to carry out semantic clustering of all of the rows into Hierarchical Clustering (Agglomerative) with 2 ranges. Use Silhouette rating. What’s the optimum variety of subsequent Level0 and Level1 clusters? What number of threads per Level1 cluster? Simply give me the stats for now, we’ll do the precise clustering later.

Immediate#5: Carry out hierarchical clustering:

Settle for this clustering with 2-levels. Add cluster subjects that summarize textual content column “thread_topic”. Cluster subjects ought to be as quick as potential with out shedding an excessive amount of element within the cluster which means.
– Level0 cluster subjects ~1–3 phrases.
– Level1 cluster subjects ~2–5 phrases.
Output JSON with columns: thread_id, level0_cluster_id, level0_cluster_topic, level1_cluster_id, level1_cluster_topic.

I additionally prompted to generate Streamlit code to visualise the clusters (since I’m not a JS knowledgeable 😄). Outcomes for a similar Chroma information are proven under.

I discovered this very insightful. For Chroma, clustering revealed that whereas customers have been pleased with subjects like Question, Distance, and Efficiency, they have been sad about areas comparable to Knowledge, Consumer, and Deployment.

Experimenting with Customized Embeddings

I repeated the above clustering prompts, utilizing simply the numerical embedding (“user_embedding”) within the CSV as a substitute of the uncooked textual content summaries (“user_text”).I’ve defined embeddings intimately in earlier blogs earlier than, and the dangers of overfit fashions on leaderboards. OpenAI has dependable embeddings that are extraordinarily reasonably priced by API name. Beneath is an instance code snippet the way to create embeddings.

from openai import OpenAI


EMBEDDING_MODEL = "text-embedding-3-small"
EMBEDDING_DIM = 512 # 512 or 1536 potential


# Initialize shopper with API key
openai_client = OpenAI(
   api_key=os.environ.get("OPENAI_API_KEY"),
)


# Perform to create embeddings
def get_embedding(textual content, embedding_model=EMBEDDING_MODEL,
                 embedding_dim=EMBEDDING_DIM):
   response = openai_client.embeddings.create(
       enter=textual content,
       mannequin=embedding_model,
       dimensions=embedding_dim
   )
   return response.information[0].embedding


# Perform to name per pandas df row in .apply()
def generate_row_embeddings(row):
   return {
       'user_embedding': get_embedding(row['user_thread_summary']),
   }


# Generate embeddings utilizing pandas apply
embeddings_data = df.apply(generate_row_embeddings, axis=1)
# Add embeddings again into df as separate columns
df['user_embedding'] = embeddings_data.apply(lambda x: x['user_embedding'])
show(df.head())


# Save as CSV ...

Curiously, each Perplexity Professional and Gemini 2.0 Professional typically hallucinated cluster subjects (e.g., misclassifying a query about sluggish queries as “Private Matter”).

Conclusion: When performing NLP with prompts, let the LLM generate its personal embeddings — externally generated embeddings appear to confuse the mannequin.

Clustering Throughout A number of Discord Servers

Lastly, I broadened the evaluation to incorporate Discord messages from three totally different VectorDB distributors. The ensuing visualization highlighted widespread points — like each Milvus and Chroma dealing with authentication issues.

Abstract

Right here’s a abstract of the steps I adopted to carry out semantic clustering utilizing LLM prompts:

1. Extract Discord threads.
2. Format information into dialog turns with roles (“consumer”, “assistant”).
3. Rating sentiment and save as CSV.
4. Immediate Google Gemini 2.0 flash for thread summaries.
5. Immediate Perplexity Professional or Gemini 2.0 Professional for clustering primarily based on thread summaries utilizing the identical CSV.
6. Immediate Perplexity Professional or Gemini 2.0 Professional to put in writing Streamlit code to visualise clusters (as a result of I’m not a JS knowledgeable 😆).

By following these steps, you possibly can shortly rework uncooked discussion board information into actionable insights — what used to take days of coding can now be completed in only one afternoon!

References

1. Clio: Privateness-Preserving Insights into Actual-World AI Use, https://arxiv.org/abs/2412.13678
2. Anthropic weblog about Clio, https://www.anthropic.com/analysis/clio
3. Milvus Discord Server, final accessed Feb 7, 2025
   Chroma Discord Server, final accessed Feb 7, 2025
   Qdrant Discord Server, final accessed Feb 7, 2025
4. Gemini fashions, https://ai.google.dev/gemini-api/docs/fashions/gemini
5. Weblog about Gemini 2.0 fashions, https://weblog.google/know-how/google-deepmind/gemini-model-updates-february-2025/
6. Scikit-learn Silhouette Rating
7. OpenAI Matryoshka embeddings
8. Streamlit