Update app.py

Final working version with motivational quote dataset

app.py

@@ -2,48 +2,66 @@ import gradio as gr
 from datasets import load_dataset
 from sentence_transformers import SentenceTransformer, util
 import torch
+from itertools import chain
+import random
 
 # Load dataset
 raw_dataset = load_dataset("asuender/motivational-quotes", "quotes_extended", split="train")
 dataset = list(raw_dataset)
 
-# Extract quotes, authors, and tags
+# Extract quotes, authors, and tags safely
 quotes = [item["quote"] for item in dataset]
 authors = [item["author"] for item in dataset]
-tags_list = [item.get("tags", "").split(", ") if item.get("tags") else [] for item in dataset]
+tags_list = []
 
-# Define fixed categories
-category_choices = ["inspirational", "energetic", "motivational", "grounding", "optimistic"]
+for item in dataset:
+    tags = item.get("tags")
+    if tags:
+        tags_list.append(tags.split(", "))
+    else:
+        tags_list.append([])
 
-# Load model
+# Define 5 fixed categories that actually exist in dataset
+fixed_categories = ["inspiration", "success", "life", "love", "courage"]
+
+# Load transformer model for semantic similarity
 model = SentenceTransformer("all-MiniLM-L6-v2")
+quote_embeddings = model.encode(quotes, convert_to_tensor=True)
+
+# Keep track of used quotes to avoid repetition
+used_quote_indices = set()
 
+# Recommendation function
 def recommend_quote(mood_input, selected_tag):
-    # Filter by selected category
-    filtered = [(q, a) for (q, a, t) in zip(quotes, authors, tags_list) if selected_tag in t]
+    # Filter quotes by tag
+    filtered = [(i, q, a) for i, (q, a, t) in enumerate(zip(quotes, authors, tags_list)) if selected_tag in t]
+
     if not filtered:
         return "😔 Sorry, no quotes found for that category."
 
-    # Unzip filtered quotes
-    f_quotes, f_authors = zip(*filtered)
+    # Filter out used quotes if possible
+    unused = [item for item in filtered if item[0] not in used_quote_indices]
+    pool = unused if unused else filtered  # fallback to all if all used
+    f_indices, f_quotes, f_authors = zip(*pool)
 
-    # Compute similarity
+    # Embed input and compute similarity
     input_embedding = model.encode(mood_input, convert_to_tensor=True)
     f_embeddings = model.encode(f_quotes, convert_to_tensor=True)
     similarities = util.pytorch_cos_sim(input_embedding, f_embeddings)
-    top_k = torch.topk(similarities, k=min(3, len(f_quotes)))
+    top_k = torch.topk(similarities, k=1)
+
+    top_index = top_k.indices[0][0].item()
+    quote_index = f_indices[top_index]
+    used_quote_indices.add(quote_index)
 
-    # Build result
-    result = ""
-    for i in top_k.indices:
-        result += f"{f_quotes[i]}\n– {f_authors[i]}\n\n"
-    return result.strip()
+    return f"“{f_quotes[top_index]}”\n– {f_authors[top_index]}"
 
+# Gradio interface
 iface = gr.Interface(
     fn=recommend_quote,
     inputs=[
         gr.Textbox(lines=2, label="What's your current mood or thought?"),
-        gr.Radio(choices=category_choices, label="Select a category")
+        gr.Radio(choices=fixed_categories, label="Select a category")
     ],
     outputs="text",
     title="MoodMatch – AI Quote Recommender",