Now, let's look at some data products examples in the real world. In this lecture, we will review three specific examples of recommender data products. We will also review the decision support dashboard for wildfire behavior prediction. To recap our last video, data products are needed when we build systems or programming interfaces that depend on predictive models. For example, we can imagine creating a predictive model to predict what users shop for, use it to recommend content to users and estimate a demand for a new product. These are all predictor models that depend on data and other models. The World Wide Web is full of data products or consumer facing products that depend on data products. Netflix, Gmail, YouTube, Instagram, or Facebook are all built as a service to consumers for things like video streaming, email, communication and social networking. However, they host specific data products as a part of their like operations, like product recommendations, ranking, sorting, discovery using mission learning, and other modeling methods. A number of dating sites recommend a possible match based on the profile of the customer making the predictive model the service itself. So it's truly a data product. Let's take Recommender Systems as a data product example. In particular, movie recommendations. I can ask the question, what would Julian rate the movie Pitch Black? The data product should give me an answer based on Julian's previous choices and other people with similar choices to Julian's. If it thinks Julian was rated high, it should recommend Julian the movie. The data product here is the system that recommends. It helps reducing the choices for the user, making it easier to reach the content he or she likes. It also potentially helps the service with the Recommender System to increase its sales by targeting the right consumer. This state of product can be modeled using data related customer choices, profile, and location. The modeling challenge is to predict opinion based on these various factors. Another example is the friend recommendations and social networking. The data product here is the feature that shows as people you may know in the user's profile. The modeling, or prediction, is to quantify how likely to users are friends. It should be based on data related to other friendships on the side, and after two users, raising the combination a factors regarding these friendships. For example, geographical proximity, or having attended the same school, or had some same friends. As an added specific example of Recommender System based data products, let's focus on showing advertisements to the consumers. These products, in general, model if a consumer would click on an ad he or she's shown. Here, recall the search term, the query, and the ad links are the output of the recommended system based on that query. An example of a data product that is dear to my heart comes from one of my projects, the Wildfire Modeling Project I mentioned when I introduced myself earlier. Through its fire map web interface, the Wildfire Project prepares many real time and archived data sets on the fly, and generate predictive maps of the fire behavior, meaning where the fire is going and at what speed. The maps can be used for decision support during the response to a fire. Using the same data, the Wildfire Project also offers services for API-based access to data, as a data product. While the underlying cyber infrastructure for these products are the same, the users and the interface to the data product are different. To summarize, in this lecture, we overviewed recommender systems as a type of data product deployed as a part of most web-based systems. We also summarized a decision-support data product that builds on predictive wildfire behavior modeling. Next, we provide more information about how recommender systems work, as recommender systems will be the overarching example in this course.
