recommendation_engine.py

@@ -43,8 +43,9 @@ class Engine:
         else:
             self.state = {
                 'channels': {
-                    'globalKeywords': {'locked': False, 'value': 8},
+                    'globalKeywords': {'locked': False, 'value': 7},
-                    'userKeywords': {'locked': False, 'value': 8}
+                    'userKeywords': {'locked': False, 'value': 7},
+                    'screenings': {'locked': True, 'value': 2}
                 },
                 'globalKeywords': {},
             }
@@ -54,11 +55,6 @@ class Engine:
                 'nextPlaylist': {'locked': False, 'value': 4},
                 'staySame': {'locked': False, 'value': 8}
             }
-        if 'userKeywordsWeights' not in self.state:
-            self.state['userKeywordsWeights'] = {
-                'themeTags': {'locked': False, 'value': 0.3},
-                'characterTags': {'locked': False, 'value': 0.7}
-            }
         self.update_keywords()
 
     @property
@@ -201,12 +197,12 @@ class Engine:
             }
 
 
-
+# NOTE for future improvement: vids_exclude element unit could be clip or in/out time pairs, rather than playlist. 
+# The same playlist could be played in the grid view as long as these are differenct clips or separate times.
     def get_recommendations(self, user, vids_exclude = []):
         channels = {k: v.get('value', 0) for k, v in self.state['channels'].items()}
         sliders = {k: v.get('value', 0) for k, v in self.state['globalKeywords'].items()}
         gridChange = {k: v.get('value', 0) for k, v in self.state['gridChange'].items()}
-        userKeywordsWeights = {k: v.get('value', 1) for k, v in self.state['userKeywordsWeights'].items()}
 
         # Exclude playlists from the most recent grid 
         playlists = copy.deepcopy(self.playlists)
@@ -215,26 +211,13 @@ class Engine:
                 if playlist["name"] in vids_exclude:
                     playlists.remove(playlist)
 
-        # For each playlist, compute user keyword score by theme and character tags
+        # For each playlist, compute user keyword score
-        user_keywords = copy.deepcopy(user.get('keywords', {}))
+        user_keywords = user.get('keywords', {})
-        theme_tags = {k.lower():v for k,v in user_keywords.items() if not k.isupper()}
-        character_tags = {k:v for k,v in user_keywords.items() if k.isupper()}
-        # manually modify some of the user keywords to match the playlist tags
-        theme_tags["god"] = theme_tags.get("god - gods",0)
-        theme_tags["visionary"] = theme_tags.get("visionary - enlightenment",0)
-        theme_tags["enlightenment"] = theme_tags.get("visionary - enlightenment",0)
-        character_tags["FEDOR MIKHAILOVICH SOFRONOV"] = character_tags.get("FYODOR MIKHAILOVICH SOFRONOV",0)
-        character_tags["SHKABARNYA OLGA SERGEEVNA"] = character_tags.get("OLGA SERGEEVNA SHKABARNYA",0)
-        character_tags["VICTORIA OLEGOVNA SKITSKAYA"] = character_tags.get("VIKTORIA OLEGOVNA SKITSKAYA",0)
-        
         score = {}
         for playlist in playlists:
-            score[playlist['name']] = random.random() * 0.001
+            score[playlist['name']] = random.random()
-            for tag in playlist['tags']:
+            for tag in [tag for tag in playlist['tags'] if tag in user_keywords]:
-                if tag in theme_tags:
+                score[playlist['name']] += user_keywords[tag]
-                    score[playlist['name']] += theme_tags[tag] * userKeywordsWeights["themeTags"]
-                elif tag in character_tags:
-                    score[playlist['name']] += character_tags[tag] * userKeywordsWeights["characterTags"]
         # Select highest scoring playlists
         playlists = sorted(
             playlists,
@@ -253,7 +236,23 @@ class Engine:
             playlists,
             key=lambda playlist: -score[playlist['name']]
         )
-        videos += playlists[:16 - channels['userKeywords']]
+        videos += playlists[:channels['globalKeywords']]
+        playlists = playlists[channels['globalKeywords']:]
+        # Count products the user has seen
+        count = defaultdict(lambda: 0)
+        for event in user.get('events', []):
+            if event.get('data', {}).get('product'):
+                count[event['data']['product']] += 1
+        # For each product in playlist tags, increment score by count
+        for playlist in playlists:
+            score[playlist['name']] = random.random()
+            for tag in set(playlist['tags']) & set(count):
+                score[playlist['name']] += count[tag]
+        # Select highest scoring playlists
+        videos += sorted(
+            playlists,
+            key=lambda playlist: -score[playlist['name']]
+        )[:16 - channels['userKeywords'] - channels['globalKeywords']]
         # Shuffle playlists (randomize layout) and shift clips (randomize start)
         random.shuffle(videos)
         return [{