How to Set Up FlutterFlow with a Content Recommendation Algorithm

Create a user_interactions collection with fields: userId (String), contentId (String), type (String: 'view', 'like', 'save', 'share'), timestamp (Timestamp), contentCategory (String, denormalized for fast queries). Add interaction logging to your app: on content page load, create a 'view' interaction. On like button tap, create a 'like' interaction. On save/bookmark tap, create a 'save' interaction. On share action, create a 'share' interaction. Also update the user document with a likedCategories array field that tracks unique categories the user has interacted with (append on like, deduplicate).

As a quick starting point before the Cloud Function pipeline, create a Backend Query on the content collection using arrayContainsAny with the current user's likedCategories array. Order by a popularity or recency field. Exclude content the user has already viewed by maintaining a viewedContentIds array in App State (populated on app load from recent user_interactions where type == 'view'). Display results in a 'Recommended for You' horizontal ListView on the home page. Each item is a ContentCard Component showing: thumbnail Image, title Text, category Badge, and a 'Recommended' label.

Create a scheduled Cloud Function generateRecommendations that runs daily. For each active user: 1) Query their user_interactions from the last 30 days. 2) Calculate a category preference score: likes=3 points, saves=2, shares=2, views=1 per interaction type. 3) Rank categories by score. 4) For the top 3 categories, query content the user has not yet viewed, ordered by popularity. 5) Score each content item: category match score + freshness bonus (newer = higher) + popularity factor. 6) Write the top 20 recommendations to users/{uid}/recommendations subcollection with: contentId, score, reason (e.g., 'Because you liked [category]'). Clear old recommendations before writing new ones.

1// Cloud Function: generateRecommendations (scheduled daily)
2const admin = require('firebase-admin');
3const db = admin.firestore();
4
5exports.generateRecommendations = async () => {
6  const thirtyDaysAgo = new Date();
7  thirtyDaysAgo.setDate(thirtyDaysAgo.getDate() - 30);
8
9  const usersSnap = await db.collection('users').get();
10  for (const userDoc of usersSnap.docs) {
11    const userId = userDoc.id;
12    const interactions = await db.collection('user_interactions')
13      .where('userId', '==', userId)
14      .where('timestamp', '>=', thirtyDaysAgo)
15      .get();
16
17    const scores = {};
18    const weights = { like: 3, save: 2, share: 2, view: 1 };
19    const viewedIds = new Set();
20
21    interactions.docs.forEach(doc => {
22      const d = doc.data();
23      const cat = d.contentCategory;
24      scores[cat] = (scores[cat] || 0) + (weights[d.type] || 1);
25      if (d.type === 'view') viewedIds.add(d.contentId);
26    });
27
28    const topCats = Object.entries(scores)
29      .sort((a, b) => b[1] - a[1])
30      .slice(0, 3)
31      .map(e => e[0]);
32
33    const content = await db.collection('content')
34      .where('category', 'in', topCats)
35      .orderBy('popularity', 'desc')
36      .limit(50)
37      .get();
38
39    const recs = content.docs
40      .filter(d => !viewedIds.has(d.id))
41      .slice(0, 20)
42      .map(d => ({
43        contentId: d.id,
44        score: scores[d.data().category] || 0,
45        reason: `Because you liked ${d.data().category}`,
46      }));
47
48    // Write recommendations
49    const recsRef = db.collection(`users/${userId}/recommendations`);
50    const batch = db.batch();
51    const oldRecs = await recsRef.get();
52    oldRecs.docs.forEach(d => batch.delete(d.ref));
53    recs.forEach(r => batch.set(recsRef.doc(), r));
54    await batch.commit();
55  }
56};

On your HomePage, add a 'Recommended for You' section. Use a horizontal ListView with a Backend Query on users/{currentUser.uid}/recommendations ordered by score descending. Each item queries the content document by contentId (using a second Backend Query or a Document Reference) to get the full content details. Display each as a ContentCard: thumbnail Image, title Text, category Badge Container, and a reason Text below (e.g., 'Because you liked Technology'). Add a 'See All' TextButton that navigates to a full recommendations page with a vertical ListView. If the recommendations subcollection is empty (new user), fall back to a 'Trending' query showing the most popular content.

Create a PreferencesPage where users can explicitly set content preferences. Display all content categories as ChoiceChips. Pre-select categories from the user's likedCategories array. Let users add or remove categories. Save changes to the user document. Also add feedback to recommendations: on each recommended item, add 'Not Interested' and 'Show More Like This' options. 'Not Interested' creates a dismissed_recommendations document. 'Show More Like This' creates a 'like' interaction boosting that category. The Cloud Function reads dismissed_recommendations to exclude those items from future results.