<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Illinois Tech ACM - Medium]]></title>
        <description><![CDATA[A student-run ACM chapter at Illinois Institute of Technology working on teaching members new technical skills, meeting technology companies and having fun building exciting projects with other students. - Medium]]></description>
        <link>https://medium.com/acm-iit?source=rss----607a3f5a4928---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Illinois Tech ACM - Medium</title>
            <link>https://medium.com/acm-iit?source=rss----607a3f5a4928---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 04:36:08 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/acm-iit" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Reversing Real Computer Scientist Syndrome]]></title>
            <link>https://medium.com/acm-iit/reversing-real-computer-scientist-syndrome-455562f98278?source=rss----607a3f5a4928---4</link>
            <guid isPermaLink="false">https://medium.com/p/455562f98278</guid>
            <category><![CDATA[imposter-syndrome]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[graduation]]></category>
            <dc:creator><![CDATA[Vinesh Kannan]]></dc:creator>
            <pubDate>Sun, 19 May 2019 00:55:31 GMT</pubDate>
            <atom:updated>2019-05-19T00:55:52.570Z</atom:updated>
            <content:encoded><![CDATA[<p><em>From the Illinois Institute of Technology Commencement 2018–2019. Full text of the undergraduate student speech for the College of Science, Department of Computer Science. Video recording below (15:13 - 27:00).</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F_8uMkE3ECKs%3Fstart%3D913%26feature%3Doembed%26start%3D913&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D_8uMkE3ECKs&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F_8uMkE3ECKs%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/c327e5b12bc5c39264df2777d9ae8bda/href">https://medium.com/media/c327e5b12bc5c39264df2777d9ae8bda/href</a></iframe><p>I’m just gonna come out and say it: I don’t like sixth graders.</p><p>They’re the worst.</p><p>I understand that you are graduating with bachelor’s degrees, master’s degrees, and PhDs, but if it’s okay with you, I’d like to take a moment to talk about sixth graders.</p><p>Sixth graders can be merciless bullies. Some sixth graders psychologically attack their friends and classmates because they think it makes them look cool. And they don’t even have Twitter accounts yet!</p><p>Of course, I am reflecting on a specific incident with some sixth graders. Four years ago, I was volunteering at a middle school, assisting a first-time computer science teacher. The students were coding with Scratch, which is an approachable and creative software that allows students to build their own projects with block-based programming. One of the students programmed a game that asked players to guess the name of a country. When the player clicked on the country, its name would appear. To be clear, this is one of the sixth graders I actually like. The student told me that she got the idea for her game during the geography unit of her social studies class. I was impressed that she translated an idea from her head into code. I was even more impressed that she persisted over several class periods until the game was working.</p><p>I told the student, “this is a great computer science project!” And the sixth grader sitting next to her chimed in: “This is just Scratch, real computer scientists use Java.”</p><p>That’s the kid I don’t like.</p><p>What do you want me to say? “Good job on your original and creative expression of computer science, but come over here and let us talk about the meaning of ‘public static void main.’”</p><p>I think the worst part of the Scratch versus Java debacle is that if these students reach a university programming course, they will hear from classmates, professionals, or even professors, that even Java does not pass this mythical bar of “real computer science.”</p><p>Many of you are already acutely aware of this effect: an anxious voice speaking in your head, trivializing your work as not yet worthy of the label “computer science.” That voice was not always there. Some demoralizing interaction, no matter how implicit, placed it there. This is our field’s version of imposter syndrome: the fear that you are not good enough, just waiting to be exposed by someone more experienced.</p><p>This “real computer scientist syndrome” extends from sixth graders all the way up to senior engineers and researchers. It might even start at a much younger age. I have seen baby toys that look suspiciously like the Towers of Hanoi problem. After all, how could a child ever grow up to be a computer scientist without understanding tail recursion at the age of three?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1001/0*mwq35-JX6I2rBIAN.jpg" /><figcaption>Amazon: VolksRose Large Creative 8 Rings Tower of Hanoi Wooden Logic Brain Teaser Puzzle for Your Kids</figcaption></figure><p>“Real computer scientist syndrome” is more than domain-specific bullying, it descends from a cognitive bias known as survivorship bias. In World War II, the Allied forces analyzed their aircraft to determine how to improve their defenses for battle. Their idea was to reinforce the areas of planes that had sustained the most damage: the wings and the fuselage. Mathematician Abraham Wald pointed out the major flaw with this plan: the Allies could only the study the planes that returned from war. Their analysis overlooked extra defenses for the cockpit and engine because fighters who took hits to those areas never made it home.</p><p>Similarly, to understand the impact of “real computer scientist syndrome,” we must pay attention to who <em>is not</em> in the room with us today: those who are not here graduating with degrees in computer science.</p><p>Class of 2019, this is our opportunity to reflect on the narrative of what makes someone a “real computer scientist,” so that it does not harm us if others try to demean our work in the future and so that we do not inadvertently pass it on to our teammates and mentees.</p><p>I have collected some examples of the “real computer scientist syndrome.” To collect these examples, I had to visit some of the most dangerous places on Earth: mostly Twitter and Quora. Some of these statements will sound absurd and seem obviously backwards. But you have been exposed to them before. More importantly, the students who will grow up to be your collaborators are hearing them right now.</p><p>Here is a taste of “real computer scientist syndrome:”</p><blockquote>A real computer scientist would<br>Never look up code on StackOverflow<br>Don’t listen to people who say<br>Skills other than programming are important<br>Java is interesting, but<br>Can get annoying after a while<br>A rockstar coder is a person who<br>Can tackle really hard problems<br>Collaborating with people from other disciplines means you<br>Have not studied enough computer science<br>People who say they can build something overnight<br>Work hard and don’t sleep until the job is done<br>You will not be the best version of yourself if you<br>Do not know how to implement a red-black tree on a whiteboard<br>No one really cares if you<br>Explore interests outside of computer science<br>To show your personality, you should<br>Spend nights and weekends on machine learning projects<br>Not everyone is able to<br>Implement a perfect dynamic memory allocator<br>Your professors don’t know how to<br>Stop you from procrastinating<br>Graduating from college does not<br>Prove that you are intelligent<br>Be cautious of people who make you<br>Document your work<br>Other engineers will look up to you if you<br>Commit many lines of code<br>Your work can be groundbreaking even if you do not<br>Make sure that others can understand<br>Your grades, your grad school, and your salary<br>Determine the kind of person you will be, not<br>The experiences you shared with your classmates and professors</blockquote><p>Thankfully, these statements are like a linked list. We can reverse them in linear time. Here are the exact same lines, but in reverse order:</p><blockquote>The experiences you shared with your classmates and professors<br>Determine the kind of person you will be, not<br>Your grades, your grad school, and your salary<br>Make sure that others can understand<br>Your work can be groundbreaking even if you do not<br>Commit many lines of code<br>Other engineers will look up to you if you<br>Document your work<br>Be cautious of people who make you<br>Prove that you are intelligent<br>Graduating from college does not<br>Stop you from procrastinating<br>Your professors don’t know how to<br>Implement a perfect dynamic memory allocator<br>Not everyone is able to<br>Spend nights and weekends on machine learning projects<br>To show your personality, you should<br>Explore interests outside of computer science<br>No one really cares if you<br>Do not know how to implement a red-black tree on a whiteboard<br>You will not be the best version of yourself if you<br>Work hard and don’t sleep until the job is done<br>People who say they can build something overnight<br>Have not studied enough computer science<br>Collaborating with people from other disciplines means you<br>Can tackle really hard problems<br>A rockstar coder is a person who<br>Can get annoying after a while<br>Java is interesting, but<br>Skills other than programming are important<br>Don’t listen to people who say<br>Never look up code on StackOverflow<br>A real computer scientist would</blockquote><p>I am grateful to all of the friends, family, and faculty who have helped us reach graduation. The class of 2019 has worked hard to reach this point, but you were all computer scientists even before you earned your degree.</p><p>You are not a computer scientist because of the algorithms you know or the programming languages you work in…</p><ul><li>You are a computer scientist anytime you creatively transform a hard problem into a version that you know how to solve.</li><li>You are a computer scientist anytime you think adversarially to find and fix flaws in a system.</li><li>You are a computer scientist anytime you create solutions for both machines and humans.</li></ul><p>The things that traditionally make us computer scientists are not the only things we have to give to the world. They may be the characteristics that the media, society, and even our institution will focus on, but we bring much more.</p><p>Many of the supposedly “real computer scientists” have behaved like sixth graders. I look forward to seeing all the ways that the class of 2019 will change our field. We have the opportunity to reverse the syndrome.</p><p><em>Thank you to the College of Science for supporting the classes of 2018 and 2019. Thank you to Michael Saelee, Robert Ellis, and Rakesh Chatrath for their feedback on this speech.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=455562f98278" width="1" height="1" alt=""><hr><p><a href="https://medium.com/acm-iit/reversing-real-computer-scientist-syndrome-455562f98278">Reversing Real Computer Scientist Syndrome</a> was originally published in <a href="https://medium.com/acm-iit">Illinois Tech ACM</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ethics of the Explore vs Exploit Tradeoff]]></title>
            <link>https://medium.com/acm-iit/ethics-of-the-explore-vs-exploit-tradeoff-7f57809f9543?source=rss----607a3f5a4928---4</link>
            <guid isPermaLink="false">https://medium.com/p/7f57809f9543</guid>
            <category><![CDATA[education]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Vinesh Kannan]]></dc:creator>
            <pubDate>Tue, 09 Apr 2019 18:31:22 GMT</pubDate>
            <atom:updated>2019-04-09T18:31:22.240Z</atom:updated>
            <content:encoded><![CDATA[<p>My statistics professor invited me to talk about ethical machine learning to her graduate statistical learning class. I picked a topic that would be technically appealing to the students and help introduce ethical concepts.</p><p><strong>Materials:</strong> <a href="https://teacher.desmos.com/activitybuilder/custom/5cabbaf742f1600ab3831c1b">Activity</a> and <a href="https://docs.google.com/presentation/d/1LeD3ra0IfvAqsOKjYMWeR7QNHjQuEeONft4oW11d0PE/edit?usp=sharing">Slides</a></p><p><strong>Time:</strong> 45–60 minutes</p><p><strong>Learning Objectives:</strong></p><ul><li>Students will analyze how their classmates estimate the potential benefits and harms of an experiment.</li><li>Students will recognize the rights of users in an experiment.</li><li>Students will evaluate multi-armed bandits as a tool for designing ethical experiments.</li></ul><h3>Explore vs Exploit</h3><p>I designed a session about the explore/exploit tradeoff, which describes the tension between wanting to choose the best action according to your data (exploit) and wanting to experiment with potentially suboptimal actions to find out if they might be better (explore).</p><blockquote>Should we give some people a potentially inferior treatment in order to learn about its effects?</blockquote><p>Academic researchers might use randomized control trials (RCTs) to determine if a medical treatment is effective. Technology companies might use A/B tests to improve the click-through rate of a website. In both types of experiments, users are split into random groups and assigned different variations of an experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*PEoDlg8JH1J7ec8BwSieYA.png" /><figcaption>Example of an Amazon A/B test. Credit: Wolfgang Bremer, via Medium</figcaption></figure><p>Researchers who work with human subjects are bound by internal review boards (IRBs) and The Federal Policy for the Protection of Human Subjects, known as “the Common Rule.” Users must give consent to participate in such experiments. Companies obtain consent when users agree to their terms of service, but users may not be fully aware of what they have agreed to.</p><p>Actions can be legal, but not ethical. Studies can be statistically-sound, but not ethical. I wanted to introduce students to language that would help them describe to others the ethics of an explore/exploit situation.</p><h3>Case Study: Newsfeed Experimentation</h3><p>I choose <a href="https://www.pnas.org/content/111/24/8788.full">a 2014 Facebook experiment</a> as a case study:</p><blockquote>“In an experiment with people who use Facebook, we test whether emotional contagion occurs outside of in-person interaction between individuals by reducing the amount of emotional content in the News Feed. When positive expressions were reduced, people produced fewer positive posts and more negative posts; when negative expressions were reduced, the opposite pattern occurred. These results indicate that emotions expressed by others on Facebook influence our own emotions, constituting experimental evidence for massive-scale contagion via social networks. This work also suggests that, in contrast to prevailing assumptions, in-person interaction and nonverbal cues are not strictly necessary for emotional contagion, and that the observation of others’ positive experiences constitutes a positive experience for people.”</blockquote><blockquote>- Kramer, Guillory, Hancock. 2014. Emotional contagion through social networks.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*bE48Lk_q4cHldk5WwQ3WDQ.png" /><figcaption>Headlines showing media outcry about Facebook’s newsfeed emotions experiment. Image: CBS News (2014).</figcaption></figure><p>I designed <a href="https://teacher.desmos.com/activitybuilder/custom/5cabbaf742f1600ab3831c1b">an activity on Desmos</a> to walk students through the steps of the <a href="https://www.scu.edu/ethics/ethics-resources/ethical-decision-making/a-framework-for-ethical-decision-making/">ethical decision making framework</a> from the Markkula Center for Applied Ethics at Santa Clara University.</p><ol><li>Recognize an Ethical Issue</li><li>Get the Facts</li><li>Evaluate Alternative Actions</li><li>Make a Decision and Test It</li><li>Act and Reflect on the Outcome</li></ol><p>The Markkula Center outlines five sources of ethical standards. For this activity, I choose to focus on two: utilitarian ethics and rights ethics.</p><h3>Utilitarian Test: Bentham’s Criteria</h3><p>We started with a simple utilitarian test:</p><blockquote>What could be some benefits from this experiment? List at least two.</blockquote><blockquote>What could be some harms from this experiment? List at least two.</blockquote><blockquote>On the whole, is this experiment more beneficial or more harmful?</blockquote><p>In my experience, engineers and data scientists commonly talk about controversial decisions in terms of potential benefits and harms. However, if a coworker or manager does not agree with their assessment of how impactful or how likely those outcomes are, they may dismiss the concerns.</p><p>I asked students to analyze their classmates’ responses and figure out which of <a href="http://caae.phil.cmu.edu/cavalier/80130/part1/sect4/BenandMill.html">Bentham’s Criteria</a> they were using:</p><ul><li><strong>Intensity:</strong> How strong is the benefit or harm?</li><li><strong>Duration:</strong> How long will the benefit or harm last?</li><li><strong>Certainty:</strong> How likely is the benefit or harm to transpire?</li><li><strong>Nearness:</strong> How soon might the benefit or harm be felt?</li></ul><p>As I had hoped, the exercise generated disagreement. Students noticed that their classmates did not use all four dimensions. Students struggled to quantify and compare the benefits and harms that others mentioned.</p><h3>Rights Test: Kant’s Mere Means Principle</h3><p>Statements such as “this could cause irreversible psychological damage” and “all humans have a right to consent” did not fit cleanly into the criteria. To contrast utilitarian ethics with rights ethics, I introduced the class to Kant’s Mere Means test:</p><blockquote>Are the subjects being used merely as a means to the company’s end (goal)?</blockquote><blockquote>(1) In principle, could the subjects consent to the course of action?</blockquote><blockquote>(2) Is the company contributing to their users’ end (goal)?</blockquote><p>Kant argues that to be <em>just</em>, we ought to avoid using others as mere means. To be <em>beneficent</em>, we must sometimes also contribute to others’ goals. For more discussion of the mere means principle, read <a href="http://people.morrisville.edu/~galuskwj/oneill.html">Onora O’Neill’s article</a> or <a href="https://philosophy.stackexchange.com/questions/51375/applying-the-mere-means-principle">this Philosophy StackExchange post</a>.</p><ul><li>Students found it easy to assign values to Bentham’s criteria, but still disagreed with each other over their conclusions.</li><li>Most students concluded that users could not reasonably consent to this experiment, but were unsure whether or not it advanced their ends.</li></ul><p>Using multiple sources of ethical standards can bring more clarity to a decision. Most students find it easy to start talking about utility, but the mere means test provides a framework for discussing rights.</p><h3>Will Multi-Armed Bandits solve explore/exploit?</h3><p>With some ethical discussion under their belts, I introduced the students to a class of algorithms that have been billed as an alternative to A/B tests.</p><p>The Multi-Armed Bandit (MAB) randomly chooses between exploration and exploitation, in hopes of minimizing the number of users who are exposed to an inferior treatment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*Z3PI-CVAvP6PWdifCgqSdg.png" /><figcaption>Each time the multi-armed bandit has to make a decision, it either explores, by trying a random action, or exploits, by trying the action it current believes to be the best.</figcaption></figure><p><a href="http://cs-blog.khanacademy.org/2014/03/ab-testing-curriculum-to-sneak-peek-or.html">In this blog article</a> from 2014, the Khan Academy Computing team describes one of their A/B test experiments about motivational content for learners. A commenter wonders if MAB might improve the results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cibeiu5GJ5j5duO20moLfg.png" /><figcaption>A commenter on <a href="http://cs-blog.khanacademy.org/2014/03/ab-testing-curriculum-to-sneak-peek-or.html">the Khan Academy Computing blog</a> wonders if bandit algorithms can predict whether or not a video sneak peak will motivate a student.</figcaption></figure><p>Engineers, data scientists, marketers, and managers who want better, faster results may be tempted to engage in the technical debate. <a href="https://vwo.com/blog/multi-armed-bandit-algorithm/">Paras Chopra from VWO</a> and <a href="http://stevehanov.ca/blog/index.php?id=132">Steve Hanov</a> presented their opposing arguments for AB testing vs multi-armed bandits applied to website optimization.</p><p>MAB is called a “greedy algorithm,” because it makes decisions based on the local optimum. This means that it is not guaranteed to find the “true” best option. MAB captures the explore/exploit tradeoff because if data scientists choose too low of an exploration rate, MAB can lock in on whichever option initially gives it the most rewards. With too high an exploration rate, MAB will continue to test potentially suboptimal options. Students who program in Python can play with variations of MAB in <a href="https://www.kaggle.com/vingkan/multi-armed-bandit">this Kaggle kernel</a>.</p><p>I have noticed some other variations of MAB in machine learning and reinforcement learning papers that should make data scientists reflect:</p><ul><li><strong>Upper Confidence Bound 1 (UCB1)</strong> chooses the option that has the highest upper confidence bound for reward, even if variation is high. As explained by <a href="https://jeremykun.com/2013/10/28/optimism-in-the-face-of-uncertainty-the-ucb1-algorithm/">Jeremy Kun</a>, this algorithm minimizes “regret,” a measure of missed rewards. Could early users receive extreme options?</li><li><strong>Contextualized Bandits (CMAB)</strong> use attributes about the user and the option to estimate reward. <a href="http://rob.schapire.net/papers/www10.pdf">Researchers from Yahoo! and Princeton</a> used contextualized bandits to personalize news articles on the Yahoo! front page. Could some of these attributes be inappropriate?</li><li><strong>Behavior Constrained Contextual Bandits (BCCB)</strong> can learn rules about what options are permissible for certain users. <a href="https://www.ijcai.org/proceedings/2018/0843.pdf">IBM researchers</a> trained a BCCB for movie recommendations to avoid showing mature content to young viewers. What could happen when bandits have to optimize both rewards and behavioral constraints?</li></ul><p>I argued to the class that while bandit approaches may reduce the number of people exposed to suboptimal treatments, <strong>they do not change the rights of the users involved in the experiment.</strong></p><ul><li>How robust are these bandit algorithms to scenarios where users are allowed to opt-out of the experiment?</li><li>What is lost when minimizing regret and maximizing reward?</li><li>Are people who use bandits treating their users as mere means?</li></ul><p>I am not convinced that better bandit algorithms alone will make experiments more ethical. Even if the students disagree with my argument about bandits, I hope this session gave them new perspectives for discussing the ethics of the explore/exploit tradeoff. Every team should be able to discuss whether or not their experiment is ethical.</p><p><em>Thank you to Dr. Lulu Kang for her support and encouragement.</em></p><p><a href="https://github.com/vingkan"><em>Vinesh Kannan</em></a><em> studies Computer Science at the Illinois Institute of Technology. He taught a high school enrichment class about </em><a href="https://github.com/vingkan/ethicalml"><em>ethical machine learning</em></a><em> and contributed to </em><a href="https://ethicalcs.github.io/"><em>open source ethical computer science activities</em></a><em> for college courses.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7f57809f9543" width="1" height="1" alt=""><hr><p><a href="https://medium.com/acm-iit/ethics-of-the-explore-vs-exploit-tradeoff-7f57809f9543">Ethics of the Explore vs Exploit Tradeoff</a> was originally published in <a href="https://medium.com/acm-iit">Illinois Tech ACM</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction to Recommendation Systems]]></title>
            <link>https://medium.com/acm-iit/introduction-to-recommendation-systems-ac9e25064b60?source=rss----607a3f5a4928---4</link>
            <guid isPermaLink="false">https://medium.com/p/ac9e25064b60</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data-mining]]></category>
            <category><![CDATA[recommendation-system]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-analysis]]></category>
            <dc:creator><![CDATA[Erick Torres-Moreno]]></dc:creator>
            <pubDate>Tue, 26 Mar 2019 20:19:37 GMT</pubDate>
            <atom:updated>2019-03-26T20:21:37.730Z</atom:updated>
            <content:encoded><![CDATA[<p>Have you ever wondered how Amazon recommends products to you?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6_Rja_jC2xXp-WiaF0mpPQ.png" /><figcaption><a href="http://amazon.com/">amazon.com</a></figcaption></figure><p>Imagine just finishing your order for that Apple watch you’ve been saving money for and suddenly Amazon recommends you a good looking band that you didn’t know you wanted but are now suddenly inclined to buy. Or perhaps, you just finished googling about the top ten cases for your new iPhone XR and suddenly once you hop into the site the same cases seem to appear on your homepage.</p><p>Creepy? Maybe. Magic? Not really. Companies like Amazon, Spotify, Netflix, and even Facebook do this by using <strong>Recommendation Systems</strong> to help users discover items that they might like or are more likely to purchase.</p><h3>Types of Recommendation Systems</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jhRluL9LADW2d5zgAxVCpA.png" /><figcaption><a href="https://www.cameo.com/perezhilton">https://www.cameo.com/perezhilton</a></figcaption></figure><p>There are two types of Recommendation Systems that are used in practice. <em>Note: We will be using the company </em><a href="https://www.cameo.com/"><em>Cameo</em></a><em> (a company that helps users buy personalized videos from celebrities) in our examples.</em></p><p><strong>Content-Based Systems:</strong> These systems tend to look at the characteristics or attributes of items in order to compare them to items that a person or user has purchased.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/908/1*C_NYDhIC8omI4L-7xyJfDw.png" /><figcaption>We notice that a user likes a certain type of celebrity and recommend similar celebrities!</figcaption></figure><p>For example, if we were implementing Content-Based Systems in Cameo we would want to recommend Cameo’s users celebrities that they might like to purchase a Cameo from. Let us imagine for a second that you are one of the users. Using this approach we would want to compare the celebrities that you have purchased in the past with celebrities Cameo has available. This could mean comparing the movies these celebrities appeared in, the genres they categorize themselves in, or even the average ratings they are given by customers. The more similar the celebrities are then the more likely Cameo will recommend these celebrities to you.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/912/1*rqSCsQfONTl7qbN5AzsOYw.png" /><figcaption>Users that rated the same things highly might have other things in common!</figcaption></figure><p><strong>Collaborative Filtering Systems</strong>: These systems focus more on the what the users have purchased, watched, or liked/rated in the past. They will compare similar users to each other or perhaps compare items rated similarly in order to make recommendations.</p><p>For example, Cameo might use this approach by finding users that have purchased similar celebrities that you have. They might then notice that these other users have also bought videos from celebrities like Kendrick Lamar and Taylor swift but you didn’t. Therefore, Cameo might recommend you these two celebrities.</p><h3>Representing the Information</h3><p>Before we start diving deeper into recommendation systems it’s important to understand how the data is organized and used.</p><p><strong>The Utility Matrix</strong>: This matrix is very important since<strong> it contains the relationship between the users and the items</strong>. The rows represents each user and the columns each item. The entries in the matrix can be binary (1 or 0) which may represent whether a user purchased an item or not. We can also have the entries represent a rating (ex: 1–10) which would represent the rating a user gave to a particular item.</p><p>Note: The zeros in the matrix can also represent “blank” entries or in our example unrated/undiscovered celebrities. These “blank” entries are often what we want to predict or decide if a user will like the item these blank entries belong to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/655/1*6KTg1CoQSYHCjvBEi4jEIw.png" /><figcaption>Utility Matrix with Binary entries for three users and three artists</figcaption></figure><p><strong>Item Profile</strong>: This is used more in Content-Based Filtering. However, the item Profile is another matrix that contains rows for each item available and columns for the attributes or characteristics that can describe the items. In the case of Cameo, this matrix can give us information like the genre a certain celebrity belongs to or the hometown of a celebrity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/567/1*H9i6WSBHTs_kb-qyvU5bUQ.png" /><figcaption>Item Profiles for these three celebrities</figcaption></figure><p>Notice how the entries in the Item Profile vary from numbers to words. This is important because we often would like every matrix to be uniform to one type in order to compare them with each other. There are multiple ways to solve this but I would leave you with your curiosity to uncover and learn about them.</p><p><strong>User Profiles</strong>: This again is used more in Content-Based Filtering. <strong>This matrix represents the user’s preferences to attributes of an item</strong>. Similarly to the Utility Matrix, It’s rows will represents each user but the columns represent the attributes of an item instead of the items themselves. This matrix can tell us information such as what genre does a user prefer in celebrities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/497/1*pGvpbKS0eoO7nJJ7XAe6GQ.png" /><figcaption>The user profiles for two users</figcaption></figure><p>The above tells us that 20% of the celebrities User 1 are in the rap genre and 30% of them are in the latin genre. This information would be compared to that of the item profiles in order to see if there are any celebrities that match those attributes that User 1 likes.</p><h3>Similarity Measures</h3><p>So we’ve talked a lot about similar customers and similar items. But how exactly do we measure how similar these things are to eachother?</p><p>Well first, <strong>we treat each row (or sometimes columns) as a vector</strong>. For example, in the case of the Utility Matrix each vector would represent the values of each column for a user.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/804/1*nTtv8HbB2lA8u6EEW2WijQ.png" /><figcaption>Representing a utility matrix with two users and two celebrities</figcaption></figure><p>Second, <strong>we choose a similarity measure to compare the vectors</strong>. Similarity measures determine how similar vectors are by comparing the number of common items, the distances between the vectors, or even the angle that these vectors make.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/697/1*Q4xQoV8k_7S7xB-NfvFdrw.png" /><figcaption><a href="http://techinpink.com/2017/08/04/implementing-similarity-measures-cosine-similarity-versus-jaccard-similarity/">http://techinpink.com/2017/08/04/implementing-similarity-measures-cosine-similarity-versus-jaccard-similarity/</a></figcaption></figure><p>There are plenty of similarity measures to choose from but the most popular ones are <a href="http://techinpink.com/2017/08/04/implementing-similarity-measures-cosine-similarity-versus-jaccard-similarity/"><strong>Jaccard and Cosine similarity</strong></a>.</p><p>One of the reason these two similarity measures are often used is because they deal with the problem of having very sparse data. For example, because Cameo has many celebrities and people often haven’t purchased or even know about many of them their will be many 0s in our utility matrix. Therefore, the 0s in our utility matrix are not as important because they do not provide us with much information and so we want to focus more on what people have bought.</p><p>It is also important to note that although these similarity measure deal with a similar problem they do not do it in the same way and can give very different results. In fact,<strong> you can say that each similarity measurement that exists has their very own definition of “similarity”</strong>. Therefore, it is always important to understand what similarity measures you are using in your recommendation system or be ready to test multiple to see if any give improved results.</p><h3>Evaluating our Recommender System</h3><p>So now that we’ve built our awesome recommender system we need a way to measure how “correct” it is. The way we determine how good our system is will depend on how we approach our recommendation.</p><p>We can treat our recommendation as a as a ranking task or a rating task. <strong>In other words, are we recommending a set of items for each user that we think are the most important (top-K) or are we predicting the rating that a user will give to a certain item.</strong></p><p>Treating a recommendation as a rating task requires that you have data or a utility matrix where users have given some sort of ratings to items. An example would be something like movie ratings or product ratings. This will require a different type of scoring metric than if we were to do a ranking task because we want to make sure that the correct ratings are given to the items. In other words treating it like a rating task is like treating it as a prediction problem where your goal is to predict the rating. The preferred metric for this is often <strong><em>RMSE</em></strong> or <strong><em>MAE.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/524/1*DC4mfD2vplwYek--SzcEyQ.jpeg" /></figure><p>Now, when we treat our recommendation as a ranking task we often talk about <a href="https://www.youtube.com/watch?v=EeXBdQYs0CQ">Top-N recommender systems</a>. These are systems that return recommendations in order of what the user will most likely buy or like.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7hLGdO-xqBiHyM5gBKAw2w.png" /><figcaption>Amazon uses a Top-N recommender system to recommend items you are most likely to buy</figcaption></figure><p>One simple way of evaluating Top-N recommender systems is by simply looking at the hit rate. For each user, how many “hits” or correct recommendations did we make. We would usually take some average and also sometimes add weights where the first recommendations are more important then the last.</p><p>However, the most <strong>popular ranking metrics are </strong><a href="http://sdsawtelle.github.io/blog/output/mean-average-precision-MAP-for-recommender-systems.html"><strong>MAP</strong></a><strong> and </strong><a href="http://fastml.com/evaluating-recommender-systems/"><strong>NDCG</strong></a>. The use of a metric will depend on the application of your recommendation system and the data you are working with.</p><p>The MAP metric is often used when you only care about whether your recommendations are relevant or irrelvant to the user. MAP computes the mean of the Average Precision over all your users. This means that for each user you take the list or set of recommendations you gave it and compare it to a correct set. Average Precision not only rewards the fact that you got a “hit” or correct recommendation but also if you placed the more correct recommendations at the top of your set or list. You then add all the those AP scores and divide by the total number of users.</p><p><strong>Overall MAP has three main characteristics:</strong></p><ul><li>Rewards successful recommendations</li><li>Rewards recommendations that are front-loading. Meaning correct recommendations that are placed at the top.</li><li>Does not penalize you if you add non-matching recommendations to your list. However, it prefers when you front-load the correct recommendations</li></ul><p><em>Note: We did not talk about NDCG or RMSE in depth but it is important to know that these are not useful when dealing with binary data (0 or 1). These metrics are often used for non-binary data where theres a concept of relevance scores or ratings. Therefore, these metrics often use averages or exponential/logarithmic functions to compute the evaluations. These techniques don’t make sence on binary data because they do not give any useful information about it.</em></p><h3>Drawbacks</h3><p>Recommender Systems sound great! But they are not perfect. These systems can be very useful tools to enhance the customer experience and increase sales for a company because they lend themselves tom customization. However, they often have many drawbacks and may not be suitable for every situation.</p><p>One of the drawbacks comes from the fact that these systems require lots of data. What happens when you have new users? What if you are a new company? What if you don’t require user accounts? This is often known as <strong>cold start</strong> and can be a big problem when dealing with recommender systems. Often it might be the case that a recommender system may not be a great option or you may need to come up with novel solutions to solve the lack of data such as using cache data from your users or asking your users initial questions.</p><p>Another big drawback of recommender systems is that they often don’t work well with incredibly<strong> sparse data</strong>. In other words, if there are many items to be recommended, users are completly different, or there is a lot of irrelevant item attributes then it may be hard for the system to find similar vectors or patterns.</p><p>Furthermore, it can be computationally expensive in practice. If you’re Amazon and you have millions of users then comparing each user vector becomes a huge problem especially since these vectors are always changing. You may even run into some slow performance when implementing a recommendation system on data with only 1000 rows.</p><p>Pondering these issues and looking at other ways recommendation systems can go wrong is important. Therefore, I challenge you to think about other problems with recommendation systems. When would recommendation systems not be good idea? When is one type of recommendation system better than the other?</p><h3>Exploring More</h3><p>We have only scratched the surface of recommendation systems. There are many important topics that we have yet to explore.</p><p>As mentioned before we did not discuss other similarity metrics or even specific differences between Jaccard and Cosine. Understanding these metrics are crucial since they may in fact give you different recommendations and some are more appropriate in different domains then others.</p><p>We have also yet to talk about how one can use Machine Learning methods such as decision trees and clustering. Furthermore, there are also ways we can use <a href="https://medium.com/@paritosh_30025/recommendation-using-matrix-factorization-5223a8ee1f4">Matrix Factorization and Dimensionality Reduction</a> using things like SVD (singular-value decomposition) to take care of all the unimportant information such as 0s or un-discovered items. These methods can sometimes improve our systems and are interesting topics to learn about.</p><p><strong>Below are just a couple of resources for you to check out:</strong></p><p><a href="https://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf">Amazon Item-to-Item Collaborative Filtering</a></p><p><a href="https://en.wikipedia.org/wiki/Recommender_system">Wikipedia’s Recommender Systems</a></p><p><a href="https://www.netflixprize.com/assets/GrandPrize2009_BPC_BellKor.pdf">The BellKor Solution to the Netflix Grand Prize</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ac9e25064b60" width="1" height="1" alt=""><hr><p><a href="https://medium.com/acm-iit/introduction-to-recommendation-systems-ac9e25064b60">Introduction to Recommendation Systems</a> was originally published in <a href="https://medium.com/acm-iit">Illinois Tech ACM</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>