INTRODUCTION TO

COMPUTING SCIENCE AND PROGRAMMING

Lecture 7.2: Dictionaries and Files – Continue
Exploration Topic: Recommendation Systems

CMPT 120, Spring 2023, Mohammad Tayebi

 Class Agenda

• Last Time • Today

• Dictionary Operations • Recommendation Systems
• Dictionary Methods • Recommendation System Development
• Finding a File on your Disk
• Reading Files
• Writing Files

• Reading
• cspy ch. 11 & 12

CMPT 120, Spring 2023, Mohammad Tayebi 2

 Exploration Topic

Recommendation Systems
Related Course: CMPT 353

* Next slides are adopted from Subbarao Kambhampati.

CMPT 120, Spring 2023, Mohammad Tayebi 3
 Recommendation Systems
• Recommenders are instances of
personalization software.
• Personalization concerns adapting to
the individual needs, interests, and
preferences of each user.
• Music, book, restaurant, product,
content …
• From a business perspective, it is
viewed as part of Customer
Relationship Management (CRM).
• Recommenders have been shown to
substantially increase sales at on-line
stores.
CMPT 120, Spring 2023, Mohammad Tayebi 4
 Netflix Prize

• Task
• Given users ratings on some movies
• Predict user ratings on other movies
• If John rates
• “Mission Impossible”: 5 https://www.cnbc.com/

• “Over the Hedge”: 3, and

• “Back to the Future”: 4,
• Grand Prize
• $1M
• how would he rate “Harry Porter”, … ?
• 10% improvement over the existing system
• Performance • by 2011 (in 5 years)
• Error rate (accuracy)
• Participants
• 51K contestants
• 41K teams
CMPT 120, Spring 2023, Mohammad Tayebi
• 186 countries
5
 Recommendation Systems - Examples

• Pandora music recommendation

• Finding songs similar to user’s favorite songs

• YouTube video recommendation

• Suggesting videos considering users previous activities

• Netflix movie recommendation

• Offering videos that share the same characteristics with
movies that a user rated highly

CMPT 120, Spring 2023, Mohammad Tayebi 6

 Types of Recommendation Systems

• Collaborative recommendation
• Recommending a user based on the
preferences of other similar users

• Content-based recommendation
• Recommending based on item features
considering users previous actions

CMPT 120, Spring 2023, Mohammad Tayebi 7

 Content-Based Recommending
Content/Profile-based
• Recommend items to customer C similar to previous items Red
Mars

rated highly by C. Found

ation

• E.g in movie recommendation, recommend movies with same actors, Juras-

sic
Park
Machine User
director, genre and so on. Lost
World
Learning Profile

• Uses machine learning algorithms to induce a profile of the 2001 Neuro-

mancer
2010

users preferences based on a featural description of content. Differ-

ence
Engine

Advantages Challenges
• Finding meaningful features
• No need for data on other users.
• Learning users’ tastes as a function of these content
• Able to recommend to users with unique tastes features
• Able to recommend new and unpopular items • Overspecialization by not being able to recommend
outside users’ profiles
• Recommendation for new users (cold starts)

CMPT 120, Spring 2023, Mohammad Tayebi 8

 Collaborative Filtering Recommendation
• Finding a set of similar users whose interests are
“similar” to a given user interests
• How to measure users' similarity?
• Jaccard index, Cosine similarity, …
• Normalize ratings and compute a prediction from a
weighted combination of the selected users’ ratings
• Present items with highest predicted ratings as
recommendations

Jaccard index Cosine similarity

CMPT 120, Spring 2023, Mohammad Tayebi 9

 Item-User Matrix

• The input to the collaborative

filtering algorithm is a matrix where
rows are items and columns are users
• Can think of users as vectors in the
space of items (or vice versa)
• Computing similarity between users
• Finding who are the most similar users
• Computing similarity between items
• Finding what are the most correlated items
https://buomsoo-kim.github.io/

CMPT 120, Spring 2023, Mohammad Tayebi 10

 This page is intentionally left blank.

CMPT 120, Spring 2023, Mohammad Tayebi 11

 Recommendation System - Implementation

• We want to develop a collaborative-filtering-based recommendation system.

• We use amazon users’ reviews for music instruments. In the table below you can see the data
columns of this dataset.
• We can first detect the most similar users to a given user C, and then recommend new items to C
based on the list of items purchased by similar users, or
• We can detect the most similar items to every item purchased by user C and then recommend new
items to user C.
• We use Jaccard index to compute the similarity.
https://s3.amazonaws.com/amazon-reviews-pds/tsv/amazon_reviews_us_Musical_Instruments_v1_00.tsv.gz
Amazon Reviews
Dataset Columns

CMPT 120, Spring 2023, Mohammad Tayebi 12

 Recommendation System Functions
• def read_dataset()
• Reading dataset from file
• def data_preparation()
• Preparing data for processing
• def similar_items()
• Finding most similar items to a given item
• def jaccard()
• Computing Jaccard index for two given lists https://medium.com/web-mining-is688-spring-2021

• def main()
• Main function to initiate the program

CMPT 120, Spring 2023, Mohammad Tayebi 13

 The file reading will be in UTF-8 to
handle non-ASCII characters in the file.
1. def read_dataset(path):
2. myfile = open(path, 'r', encoding='utf-8')
3. header = myfile.readline()
4. header = header.strip().split('\t’)
5. dataset = []
6. for line in myfile:
7. fields = line.strip().split('\t')
8. dict_fields = dict(zip(header, fields))
9. dataset.append(d)
10. return dataset

Zip function in Python is used to iterate over two lists. It takes corresponding
elements from two lists and merges them in a tuple.
import gzip
countries = ["UK", "Italy", "Canada"]
capitals = ["London", "Rome", "Ottawa"]
country_capital = zip(countries, capitals)
CMPT 120, Spring 2023, Mohammad Tayebi 14
 1. def data_preparation(dataset):
2. users_per_item = {}
3. for d in dataset:
4. user = d['customer_id']
5. item = d['product_id']
6. if item not in users_per_item:
7. users_per_item[item] = [user]
8. else:
9. temp_list = users_per_item[item]
10. temp_list.append(user)
11. users_per_item[item] = temp_list
12. return users_per_item

CMPT 120, Spring 2023, Mohammad Tayebi 15

 Sorting the list in descending order.

1. def similar_items(upi, itemid):

2. similarities = []
3. users = upi[itemid]
4. for i in upi:
5. if i == itemid:
6. continue
7. similarity = jaccard(users, upi[i])
8. similarities.append((similarity, i))
9. similarities.sort(reverse=True)
10. return similarities[:10]

CMPT 120, Spring 2023, Mohammad Tayebi 16

 • The values in a set are unique because there cannot be any duplicates.
• When we convert a list to a set, all duplicates are removed.
• The intersection() method returns a set that contains the similarity
between two sets.
• The union() method returns a set that contains all items from both of the sets.

1. def jaccard(k1, k2):

2. s1 = set(k1)
3. s2 = set(k2)
4. numerator = len(s1.intersection(s2))
5. denominator = len(s1.union(s2))
6. return numerator / denominator

CMPT 120, Spring 2023, Mohammad Tayebi 17

 1. def main():
2. path = 'music.txt'
3. data = read_dataset(path)
4. item_user = data_preparation(data)
5. query = data[2]['product_id']
6. sim = similar_items(item_user, query)
7. print(sim)

[(0.028446389496717725, 'B00006I5SD'), (0.01694915254237288, 'B00006I5SB'),

(0.015065913370998116, 'B000AJR482'), (0.014204545454545454, 'B00E7MVP3S'),
(0.008955223880597015, 'B001255YL2'), (0.008849557522123894, 'B003EIRVO8'),
(0.008333333333333333, 'B0015VEZ22'), (0.00821917808219178, 'B00006I5UH'),
(0.008021390374331552, 'B00008BWM7'), (0.007656967840735069, 'B000H2BC4E')]

CMPT 120, Spring 2023, Mohammad Tayebi 18

 Next Lecture

Midterm Exam Preparation

Pre-reading: cspy ch. 1-12

CMPT 120, Spring 2023, Mohammad Tayebi

19