JanetBackEnd/Recommender.py

56 lines
2.6 KiB
Python

import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
import random
class Recommender:
def __init__(self, retriever):
self.curr_recommendations = {}
self.recommended = {}
self.retriever = retriever
self.rand_seed = 5
def _new(self, username, material):
if username not in curr_recommendations:
return True
for row in self.curr_recommendations[username]:
if row['id'] == material['id'] and row['type'] == material['type']:
return False
return True
def _match_tags(self, username, material, interest):
score = 0.7
for tag in material['tags']:
if cosine_similarity(np.array(self.retriever.encode([tag])),
np.array(self.retriever.encode([interest]))) > score:
if self._new(username, material):
print('hi')
self.curr_recommendations[username] = self.curr_recommendations[username].append(material) if username not in self.curr_recommendations else [material]
self.recommended[username] = self.recommended[username].append(False) if username not in self.recommended else [False]
def generate_recommendations(self, username, interests, new_material):
for interest in interests:
for i, material in new_material.iterrows():
self._match_tags(username, material, interest)
def make_recommendation(self, username, name):
if len(self.curr_recommendations[username]) == 0:
return ""
to_consider = [idx for idx, value in enumerate(self.recommended[username]) if value == False]
if len(to_consider) == 0:
return ""
index = random.choice(list(range(0, len(to_consider))))
index = self.recommended[username][index]
#while self.recommended[index] == True:
# index = random.choice(list(range(0, len(self.curr_recommendations))))
recommendation = "Hey " + name + "! This " + self.curr_recommendations[username][index][
'type'].lower() + " about " + ', '.join(
self.curr_recommendations[username][index]['tags']).lower() + " was posted recently by " + \
self.curr_recommendations[username][index][
'author'].lower() + " on the catalogue. You may wanna check it out! It is titled " + \
self.curr_recommendations[username][index]['title'].lower() + ". Cheers, Janet"
# self.curr_recommendations.remove(self.curr_recommendations[index])
self.recommended[username][index] = True
return recommendation