#recommendationsystem

Want to create an AI-powered recommendation system? Start with data collection, employ machine learning algorithms like collaborative filtering or content-based filtering, and constantly iterate for accuracy. Dive deeper into the steps involved and unleash the potential of personalized recommendations.

Read more ➡️

I first learned of recommendation systems during my first literature review course at NCU. The topic was online reviews and recommender systems were one type of system connected with online reviews. I included some information in my paper on these systems that made recommendations to users. I thought they added a lot of value (were very helpful and useful to the consumer). Today, I learned of recommendation systems for music. I found several papers on systems that use different approaches to recommend music to users. How interesting! I would like to personally test these systems.

A Music Recommendation System Based On Psychotherapy By: Zhiyuan Liu, Wei Xu, Wenping Zhang, Qiqi Jiang https://www.sciencedirect.com/science/article/pii/S277256932300097X

I am all EARS: Using Open Data And Knowledge Graph Embeddings for Music Recommendations By: Niels Bertram, Jürgen Dunkel, Ramón Hermoso https://www.sciencedirect.com/science/article/pii/S0957417423008497

สำหรับบทความนี้ อยากจะพูดถึง Recommendation System in ML ตอนนี้หลายๆ คนน่าจะคุ้นชินกับการที่เราใช้ Social Media แล้วมีการแนะนำสินค้าต่างๆ ในระบบ หรือได้ยินจากร้านสะดวกซื้อหลายๆ แห่ง ว่า “customers who bought this, also bought this” จริงๆ แล้วเจ้าระบบแนะนำนี้ เป็นส่วนหนึ่งของ Algorithms ใน Data Analytics เลยนะ

Above image source from : https://www.xenonstack.com/use-cases/recommendation-system-deep-learning/

แล้วเจ้า Techniques หรือ Algorithms นี้มันดียังไง

Before the advent of the Internet, people used to get their tips and recommendations about various items and sources largely from family members, friends, trusted network and experts. With the proliferation of online services with the overwhelming amount of content and items, information overload has become a critical issue. The large amount of availability of choices, while beneficial, also creates problems for the users.

The various problems a user encounters due to information overload include time wasted in search for appropriate selection, making poor choice while there are better options available and difficulties in arriving at the right choice. Recommender systems are the tools that are aimed at alleviating the information overload related issues. Today, most users largely depend on the web service itself to provide recommendations of the items that would arguably reflect users’ personal choices. The concept of recommendations solutions or recommender systems (RS) is not new. Some early websites have used them actively and over time they have become a necessary element and an integral part of any large to medium e-commerce, social networking sites and web services.

Recommendations solutions are meant to produce a list of items or products to website users based on user’s profile, online behavior, item’s profile, click-through and various other attributes. There is a fundamental difference between a search engine and a recommender system. When the user knows or at least can guess what keywords or terms needed to apply in order to find an item, a document or a product the person is looking for, she uses a search engine. In the case of a recommender system, the list of recommendations is generated for items, documents, products, etc. without an explicit search conducted by the user.

From user’s perspective, this is wonderful, because it allows the user to discover new content corresponding to her taste that she was not aware of earlier. By providing relevant but new choices, recommender systems make user interaction with the sites interesting and significantly enrich user’s experience. Web services implement recommendations solution for multiple reasons.

Firstly, an e-commerce site would like to sell more products, or services and RS helps a site to increase its conversion rate by transforming more visitors of the site to buyers.

Secondly, RS contributes to cross selling – sale of more types of products or services to the buyer. For example, an e-commerce site of electronic products may provide the buyer of a camera a precise recommendation list of lenses, memory cards, and cases. These items, possibly, would have been difficult for the buyer to discover without the help of the RS. Thirdly, data collected from the RS can help the web service to organize and optimize its content and items and improve inventory control. And most importantly, RS increases users’ satisfaction and their loyalty. RS has the capability of consistently delivering relevant, new and serendipitous items through an effectively designed interface. It can often predict the next steps of a user successfully and provide clear calls for action. Most users highly value interaction with a website with these capacities. An extended user-experience supported by the RS in turn facilitate converting the user to a loyal customer.

Recommender systems use myriad of techniques and algorithms to calculate and deliver recommendations. In its most basic form, many websites offer a non-personalized recommendation list of most popular items. The rationale behind this approach is if an item is liked by a large number of people, there is a high probability that others may also like it. Some systems perform predictive analysis and deliver recommendations by computing and comparing the user’s assigned ratings to the items she liked or disliked with the ratings of all other users of the system. The assumption here is that if the users rate same items with similar ratings, their interests are quite congruent. It is safe to recommend an item that the user did not rate but others with similar interest have rated it highly. Another method is to elicit recommendations to the user of an item based on what others have bought after searching or purchasing the same item. In more sophisticated RS extensive knowledge about the users, items, context and various other complex attributes are used.

Often, the systems are dynamic and consider the preferences of the user in question in real time. Rating of an item ascribed by a user is the most commonly used data in recommender systems. Often, web services consider that ratings are the most important indicator for a recommender system. In reality, this is not the case. The reason is people feel necessary to ascribe a rating to an item when it induces a strong positive or negative emotional link. People without any strong opinion about the item tend not to rate anything.

As, usually, small percentage of the population prefer to rate items, for many items the quantity of ratings are not representative enough. As a result, these items are difficult to cover in the computation of a recommendation list. So, although the ratings are one of the good indicators for recommendations it should not be the only one. One method of compensating this problem is to observe user’s behavior. If a customer purchased a product or watched a movie, and the system has the ability to monitor the behavior of a large number of customers, the data collected from this produces useful clues, which can become a viable attribute for the recommender algorithm.

Recommender systems today are highly sophisticated technology that applies machine learning, statistical methods, artificial intelligence agents and other state-of-the-art tools. Web services interested in implementing RS to their system should analyze, compare and evaluate the algorithms and systems and select the one suit their purpose best.

As mentioned earlier, Recommender systems increase revenue for a web service in multiple ways. They help clients of the web service to discover items that they would like to purchase. They facilitate conversion of visitors into clients, increase cross-sell by recommending complementary items and bolster development of a loyal customer base by improving the user-experience.

Original source: http://trouvus.com/recommendations-solutions-for-web-services/

Customer engagement is a goal for virtually any business with an online presence (i.e. virtually any business these days). So it follows that businesses want followers, likes and the like. But what happens when the people following you on social media aren’t real? And just how widespread is the phenomenon of fake followers?

11.2% of Facebooks 1.23 billion monthly users are fake (Facebook).

14.2% of large US retailers Twitter followers are fake (Entrepreneur.com).

Google is now periodically auditing YouTube views to verify that they’re genuine (Google).

Fake Followers Can Lead to Real Results

It’s usually easier for something that is already popular to gain greater popularity than something that is completely unknown. Nobody wants to eat in an empty restaurant after all. Hence, the utility of fake followers: businesses can use an initial pool of fake followers to give the impression of popularity to those (real) people that they want to engage. But does this actually work?

In his Medium article ‘(Fake) Friends with (Real) Benefits’, data scientist Gilad Lotan provides an account of his investigation into the potential fruits of buying fake Twitter followers. For the low price of five dollars, Lotan bought 4000 followers for a bot account he created that goes by the handle @gilgul. And while these followers were fake, they all had profile photos and bios of such level that a passing glance at them wouldn’t immediately reveal anything suspicious. Upon a closer inspection, it wouldn’t be difficult to discover their spurious origin.

Almost as soon as the 4000 fake Twitter followers were obtained, @gilgul’s Klout score skyrocketed. Moreover, because Klout works with Bing, @gilgul also enjoyed higher rankings in Bing’s SERPs.

An increase in the @gilgul account’s real followers was observed as well. And while the number of fake followers began to decline because Twitter regularly deletes such accounts, over the next few months the number of real followers continued to grow. Litan speculates that the considerable number of fake followers made the @gilgul account look more credible to real people and that this perceived credibility made these people more inclined to follow @gilgul.

To sum it up, Litan’s study suggests that buying fake followers can have real benefits insofar as perceived popularity can lead to real popularity.

Fake Followers Aren’t Condoned By Social Media Companies

None of the major social media companies condone fake profiles or likes. Facebook, for instance, is nowactively combating fake likes and names on its platform. This means that if you do buy fake likes or followers–even just to jumpstart your presence–your base of fake followers may very well disappear before you have a chance to build your real following as fake profiles are routinely deleted by Facebook, Twitter and other platforms.

Fake Followers Don’t Engage and Don’t Convert

If the goal of your social media marketing is to engage people whom you want to convert into leads, it will be difficult to do so with people who don’t actually exist. And when it comes to lead generation, social media is a tool few businesses can afford to ignore. Indeed, Business2Community reports that 54% of B2B marketers have produced leads using social media, with 40% of those leads generating revenue. The same report also notes that 49% of marketers report social media as the most difficult lead generation method to execute, but the benefits make social media well worth the effort.

The Takeaway

There is evidence to suggest that fake followers can help you build a real audience that you can engage. However, fake followers will not themselves, of course, improve engagement or lead generation. And while fake followers may help build a real audience, you must weigh this with the risk of contradicting the policies of social media companies as well as the ethical issue of deceiving your (real) audience. Furthermore, if your goal is to improve engagement with users, solutions called recommender systems or recommendation engines are helping media publishers, ad agencies and even the social networks themselves produce positive, dramatic results.

Recommendation Engines

A recommendation engine is a system that, as its name suggests, generates recommendations for users. For instance, a VOD site may use a recommendation engine to suggest videos to users which are similar to those which they recently viewed. However, not all recommendation engines are equally effective.