Add image handling and extraction improvements for image carousels

Author: ericahan22

backend/scraping/event_processor.py

@@ -20,6 +20,16 @@
 from utils.scraping_utils import insert_event_to_db
 
 
+def _get_all_images(post):
+    """
+    Returns all image URLs for a post.
+    """
+    images = post.get("images", [])
+    # Fallback to displayUrl if images missing
+    if not images and post.get("displayUrl"):
+        images = [post["displayUrl"]]
+    return images
+
 class EventProcessor:
     def __init__(self, concurrency=5):
         self.concurrency = concurrency
@@ -66,7 +76,6 @@ def _save_event(self, event_data, ig_handle, source_url, club_type):
     @sync_to_async(thread_sensitive=True)
     def _ignore_post(self, shortcode):
         IgnoredPost.objects.get_or_create(shortcode=shortcode)
-
     async def _process_single_post_extraction(self, post):
         """Extracts event data from a single post using OpenAI."""
         async with self.semaphore:
@@ -75,33 +84,29 @@ async def _process_single_post_extraction(self, post):
 
             return await self._extract_events(
                 post.get("caption"),
-                post.get("source_image_url"), # Using S3 URL
+                post.get("all_s3_urls"),
                 post_dt
             )
 
     async def process(self, posts_data, cutoff_date):
-        """Main entry point to process a list of raw posts."""
         logger.info(f"Processing {len(posts_data)} posts...")
-        
+
         seen_shortcodes = await self._get_seen_shortcodes()
         valid_posts = []
 
         # 1. Filter Posts
         for post in posts_data:
             url = post.get("url")
-            if not url or "/p/" not in url: continue
-            
-            # Basic validation
-            if not post.get("caption") or not post.get("displayUrl"): continue
-
-            # Date Check
+            if not url or "/p/" not in url:
+                continue
+            if not post.get("caption"):
+                continue
             post_dt = parse_utc_datetime(post.get("timestamp"))
-            if not post_dt or post_dt < cutoff_date: continue
-
-            # Duplicate Check
+            if not post_dt or post_dt < cutoff_date:
+                continue
             shortcode = url.strip("/").split("/")[-1]
-            if shortcode in seen_shortcodes: continue
-
+            if shortcode in seen_shortcodes:
+                continue
             valid_posts.append(post)
 
         if not valid_posts:
@@ -110,15 +115,29 @@ async def process(self, posts_data, cutoff_date):
 
         logger.info(f"Found {len(valid_posts)} new posts. Starting image uploads...")
 
-        # 2. Upload Images
-        upload_tasks = [self._upload_image(p.get("displayUrl")) for p in valid_posts]
-        s3_urls = await asyncio.gather(*upload_tasks)
-        for post, s3_url in zip(valid_posts, s3_urls, strict=False):
-            post["source_image_url"] = s3_url
+        # 2. Upload all images for each post (with carousel support)
+        all_image_tasks = []
+        for post in valid_posts:
+            image_urls = _get_all_images(post)
+            post["all_image_urls"] = image_urls
+            all_image_tasks.append([self._upload_image(img_url) for img_url in image_urls])
+        flat_tasks = [task for sublist in all_image_tasks for task in sublist]
+        flat_results = await asyncio.gather(*flat_tasks)
+        # Map uploaded S3 URLs back to posts
+        idx = 0
+        for post in valid_posts:
+            n_imgs = len(post["all_image_urls"])
+            post["all_s3_urls"] = flat_results[idx:idx + n_imgs]
+            idx += n_imgs
 
         # 3. Extract Events
         logger.info("Extracting event data...")
-        extract_tasks = [self._process_single_post_extraction(p) for p in valid_posts]
+        extract_tasks = []
+        for post in valid_posts:
+            extract_tasks.append(self._process_single_post_extraction({
+                **post,
+                "all_s3_urls": post["all_s3_urls"]
+            }))
         results = await asyncio.gather(*extract_tasks)
 
         # 4. Save to DB
@@ -127,15 +146,44 @@ async def process(self, posts_data, cutoff_date):
             ig_handle = post.get("ownerUsername")
             source_url = post.get("url")
             shortcode = source_url.strip("/").split("/")[-1]
-            club_type = self._get_club_type(ig_handle)
+            all_s3_urls = post.get("all_s3_urls", [])
 
             if not extracted_events:
-                # Mark as ignored if AI found nothing
                 await self._ignore_post(shortcode)
                 continue
 
+            if not isinstance(extracted_events, list):
+                extracted_events = [extracted_events]
+
+            # If 1 image is provided, but AI returned multiple event objects,
+            # merge them into a single "Weekly/Summary" event.
+            if len(all_s3_urls) == 1 and len(extracted_events) > 1:
+                base_event = extracted_events[0]
+                
+                # 1. Consolidate all dates from all events into the first event
+                combined_occurrences = []
+                for evt in extracted_events:
+                    combined_occurrences.extend(evt.get("occurrences") or [])
+                base_event["occurrences"] = combined_occurrences
+                
+                # 2. Update title/description to reflect it's a summary
+                club_name = post.get("ownerFullName") or ig_handle or "Club"
+                base_event["title"] = f"{club_name} Weekly Events"
+                base_event["description"] = (base_event.get("description") or "") + "\n\n(Condensed from multiple events)"
+                
+                extracted_events = [base_event]
+                
             for event_data in extracted_events:
-                success = await self._save_event(event_data, ig_handle, source_url, club_type)
-                if success: saved_count += 1
+                # Map the correct picture to the event.
+                image_idx = event_data.get("image_index")
+                if image_idx is not None and isinstance(image_idx, int) and 0 <= image_idx < len(all_s3_urls):
+                    event_data["source_image_url"] = all_s3_urls[image_idx]
+                else:
+                    # Fallback: Use the first image (cover) if no index specified
+                    event_data["source_image_url"] = all_s3_urls[0]
+
+                success = await self._save_event(event_data, ig_handle, source_url, self._get_club_type(ig_handle))
+                if success:
+                    saved_count += 1
 
         logger.info(f"Processing complete. Saved {saved_count} new events.")

backend/scraping/main.py

@@ -10,8 +10,8 @@
 os.environ.setdefault("DJANGO_SETTINGS_MODULE", "config.settings.development")
 django.setup()
 
-from backend.scraping.event_processor import EventProcessor
-from backend.scraping.instagram_scraper import InstagramScraper
+from scraping.event_processor import EventProcessor
+from scraping.instagram_scraper import InstagramScraper
 from django.utils import timezone
 
 from scraping.logging_config import logger

backend/services/openai_service.py

@@ -99,6 +99,7 @@ def generate_event_embedding(self, event) -> list[float]:
     def extract_events_from_caption(
         self,
         caption_text: str | None = None,
+        all_s3_urls: list[str] | None = None,
         source_image_url: str | None = None,
         post_created_at: datetime | None = None,
         school: str = "University of Waterloo",
@@ -126,14 +127,17 @@ def extract_events_from_caption(
         categories_str = "\n".join(f"- {cat}" for cat in EVENT_CATEGORIES)
 
         prompt = f"""
-    Analyze the following Instagram caption and image and extract event information if it's an event post.
+    Analyze the following Instagram caption and list of images. Extract event information if it's an event post.
 
     School context: This post is from {school}. Use this to guide location and timezone decisions.
     Current context: Today is {current_day_of_week}, {current_date}
     Post was created on: {context_day}, {context_date} at {context_time}
     Current semester end date: {semester_end_time}
 
     Caption: {caption_text}
+    
+    Images (0-indexed):
+    {self._format_image_list_for_prompt(all_s3_urls) or []}
 
     STRICT CONTENT POLICY:
     - ONLY extract an event if the post is clearly announcing or describing a real-world event with BOTH:
@@ -156,6 +160,7 @@ def extract_events_from_caption(
         "price": number or null,
         "food": string,
         "registration": boolean,
+        "image_index": integer,
         "occurrences": [
             {{
                 "dtstart_utc": string,  // UTC start "YYYY-MM-DDTHH:MM:SSZ"
@@ -165,10 +170,15 @@ def extract_events_from_caption(
             }}
         ],
         "school": string,
-        "source_image_url": string,
         "categories": list            // one or more of the following, as a JSON array of strings: {categories_str}
     }}
 
+    IMAGE MAPPING RULES:
+    - You are provided with a list of images.
+    - For each extracted event, identify which specific image contains the relevant details (e.g., date/time/location).
+    - Set "image_index" to the 0-based index of that image.
+    - Otherwise, set "image_index": 0.
+    
     OCCURRENCE RULES (CRITICAL):
     - Every event MUST include at least one occurrence with a concrete UTC start time.
     - Return ALL explicit dates and times mentioned in the post as separate entries in the occurrences array.
@@ -195,7 +205,6 @@ def extract_events_from_caption(
     - For description: Make this the caption text word-for-word. If there is no caption text, use the image text.
     - If information is not available, use empty string for strings, null for price/coordinates, and false for booleans.
     - Return ONLY the JSON array text, no extra commentary.
-        {f"- An image is provided at: {source_image_url}. If there are conflicts between caption and image information, prioritize the caption text." if source_image_url else ""}
     """
 
         try:
@@ -259,12 +268,15 @@ def extract_events_from_caption(
                     "source_image_url",
                     "categories",
                     "occurrences",
+                    "image_index"
                 ]
 
                 for event_obj in events_list:
                     for field in required_fields:
                         if field not in event_obj:
-                            if field in ["price", "latitude", "longitude"]:
+                            if field == "image_index":
+                                event_obj[field] = 0
+                            elif field in ["price", "latitude", "longitude"]:
                                 event_obj[field] = None
                             elif field in ["registration"]:
                                 event_obj[field] = False

backend/utils/scraping_utils.py

@@ -72,16 +72,15 @@ def insert_event_to_db(event_data, ig_handle, source_url, club_type=None):
     if is_duplicate_event(event_data):
         return "duplicate"
 
-    # Get club_type by matching ig_handle from Events to ig of Clubs
-    try:
-        club = Clubs.objects.get(ig=ig_handle)
-        club_type = club.club_type
-    except Clubs.DoesNotExist:
-        club_type = None
-        logger.warning(
-            f"Club with handle {ig_handle} not found, inserting event with club_type NULL"
-        )
-
+    # Only fetch if club_type wasn't passed in
+    if club_type is None:
+        try:
+            club = Clubs.objects.get(ig=ig_handle)
+            club_type = club.club_type
+        except Clubs.DoesNotExist:
+            club_type = None
+            logger.warning(f"Club {ig_handle} not found, setting club_type NULL")
+    
     create_kwargs = {
         "ig_handle": ig_handle,
         "title": title,