Recommender systems such Amazon and Netflix use collaborative filtering to automate the process of recommending products and services to consumers. The collaborative filtering systems produce personal recommendations by computing similarity between your preference and others. The idea is simple. If you need to select an item from the multitude of options with which you don’t have any experience, you will rely on the choices made by other people who have same preference and tastes as yours. Most collaborative filtering systems use the following methods:
- Record the preferences of a large group of people.
- Select group of people whose preferences are similar to yours using a similarity metric. These are the people who have same taste in things as you do.
- Recommend options to you which other people who are in the same group as you prefer or like.
This type of recommendation systems have known to really work well for books, music, movies, etc. On the other hand, they don’t work well for a number of product categories such as apparel, jewelry, and beauty products. The primary reason for this is that these systems rely on users’ actions such as purchases, rating assignments, etc to predict their preferences. This type of user input tells you what type of items they like, but it does not tell why they liked these items. The product categories such as books, music, movies, etc are one dimensional in nature. It is easy to predict what user will like by observing a couple of purchases made by him.
On the other hand, the categories such as apparel and jewelry are multi dimensional in nature. A complex analysis goes through buyers’ mind when they make such purchases. For instance, a woman may decide to buy a dress for a number of reasons such as its color, its style, its patterns, sleeve type, cut, length, and how fashionable it is. Many times the buyer doesn’t know why she liked one particular item more than other, but she has feeling that this is right item for her based on her personal experience. To determine these preferences regarding dresses for her will require hundreds of data points and which in real life don’t exist. Not only that the preference determination will require modeling each product separately. For these reasons, collaborative filtering does not work for a number of product categories.
There is another approach to personalization that uses prediction based on benefit theory.
The idea for this method is based on the intuitive fact that if you find the item for a user which intrinsically benefits them, the chances are very high they will like it and appreciate it. In other words, when people buy the products they want to maximize their benefits from it. For instance, if you find a woman a dress which looks good on her and goes well with her style, the chances are high that she will like it. The implementation of this method requires explicit characterization of the users to predict what will benefit them. This method does not require figuring out explicitly or indirectly about user preferences for various attributes of a product, but it does require asking users a few questions that can help characterize them.
MyShoppingPal.com uses this approach in its personalized search engine. When a user wants to shop for an item, the search engine asks for a few questions which help characterize the user with respect to the selected item. The information needed to evaluate the benefit is different for different product categories. For clothing items, body type, height type, and personal style is needed, while for skin care products skin-type and skin-problem related information is needed. This information about the user is used to predict what will benefit them the most. For clothing, the systems tries to find items that look good on the wearers and will go well with their style. For skin care product, it focuses on the medicinal value of the product for the user.
MyShoppingPal.com has enhanced this method with the incorporation of the most common preference attributes such as favorite brands to enhance further the value of its recommender system. Also women can shop for formal dress, halter dress, cocktail dress, evening dress from MyShoppingPal.com
The prediction based on benefit is a lot more accurate than collaborative filtering, and can be used to predict what users will like for complex product categories. The challenging part of this solution is coming up with a right and small set of questions that will help characterize users without making them feel overwhelmed with them.
MyShoppingPal.com is your best to place buy jewelry, dresses, bags, and regular and designer shoes on sale, Please try our personalized search to find your perfect item at the best price.