Netflix Schema .com



Netflix SchemaAnalysis & Suggested ImprovementsThis document presents the findings from the analysis of the existing schema of and suggests a solution which the authors believe would address a number of issues simultaneously.Daniel Cortez, Ramji Enamuthu, Sampanna Kulkarni, Adam Taplin, & Jeanna Vogt3/3/2011IntroductionNetflix is a company that generates $1.6 billion of revenue and is a household name among Television viewers in the United States and Canada. Its success depends on connecting customers with the right products in an efficient manner. The current system is strong, but lacking in some key areas. While both the graphic user interface (GUI) functionality and its supporting schema should be addressed, this report provides an analysis of the existing schema, identifying the important issues and suggesting a change we felt would have the most effect and target the most significant issues. Of particular importance to us is the user’s experience in navigating the universe of available movies to watch a movie that he prefers at that point of time. Is he able to find what he wants, when and where he expects to find it?One of the important characteristics of Netflix is that almost all of a user’s interactive experiences are through the website (). It serves as the portal which enables Netflix to serve customers effectively. The underlying metadata schema at Netflix is vital to delivering an effective user experience. In exploring the website, not only did we find easily rectifiable flaws but also a potential to apply an entity-relationship model such as the ‘Functional Requirements for Bibliographic Records’ or FRBR (Lee, 2011) with significant improvement in the user’s experience. We intended our solution to be intuitive in providing the user with what he would naturally expect to see in the behavior of Netflix. Our methodology consisted of exploring individually to compile a list of issues. Each team member rated the severity of the issues on a scale of 1-5 (with 5 being the most severe), thus leading to a numeric prioritization of the issues. The chosen issue was selected through open discussion as the one which would have the greatest impact to the experience of the user. In the following report, we first list the different issues that we discovered in our exploration. We then introduce our chosen issue and develop a solution to it based on an implementation of the FRBR model and the proper construction of an ontology. We then move on to a discussion of important results and beneficial consequences that would stem from implementing our solution on the Netflix website. Issues FoundThorough examination of the Netflix website produced the following issues, which we organized into five general categories: filtering, authority control/taxonomy, user rating, profile, and Blu Ray.Filtering IssuesThese issues address the gaps in the filtering abilities and behaviors. Netflix does not allow its users to block a group of items in which they are not interested. For example, Netflix allows the users to mark a particular episode of VeggieTales in the “Not Interested category,” but it does not allow its users to block all the VeggieTales as a group. This ability would improve the precision of recommendation results (Soergel, 2005). Also, the system lacks the ability to search or filter by decade or year of release. This classification may be interesting to some users for excluding or including results. Similarly, it filters the all time top 100 list, but does not filter the top 100 list of a particular year or a particular month. Although it displays the MPAA and Common Sense ratings for all movies, it does not support filtering movies on the basis of ratings. These features should all be included from a use-based perspective (Taylor, 2009). Lastly, there is no way for cross-filtering on the Netflix site. For example, one cannot combine the genres Military & War Dramas and Tearjerkers to return movies that fall under both of those categories. Again, this limits the precision of the system’s search results (Soergel, 2005).Search/Authority Control/Taxonomy IssuesThe following are issues of authority control and taxonomy related to Netflix’s search functionality and how it displays search results. Generally, it seems that the results of a search in Netflix are anything that matches enough letters from the search query. For instance, a search for Toy Story 3 returns two people - Tony Scott and Toby Kebbell as results 4 and 5, neither of whom have anything to do with the Toy Story franchise. Here, Netflix fails to properly discriminate and reject irrelevant results (Soergel, 2005). Additionally, a generic search for the title of a television series instead of a specific season does not return results in a logical order. For instance, a search for "24" returns, in this order - "24: Season 8", "24: Season 1", "24: Redemption", "24: Season 2," "24: Season 7," "24: Season 3," "24: Season 4," "24: Season 6," and "24: Season 5." It would make more sense to return the seasons in ascending or descending order. The schema appears to lack an association between installations of a franchise. The presence of an association would also benefit the Watch Instantly feature. It currently has associations with the episodes of a single season but there are no associations between different seasons of the same show. So it allows its users to go from one episode to the next within a season of a show, but if the schema has associated seasons it could allow its users to go on to the next season without having to go back and search for it. This represents a faulty and incorrectly created ontology (Noy & McGuinness, 2000).User Ratings IssuesThese issues relate to how the schema is applied to the user ratings system. Pages that display the movies should ideally display the average star rating given by users. Currently it just shows Netflix's best guess for its users. The user has to click on the movie and go to the details screen to see the real rating. While it’s nice that Netflix attempts to predict a user’s opinion, it’s not necessarily accurate and may mislead the user. In this, Netflix exhibits a bias—albeit a mathematically derived one—where metadata should be objective (Taylor, 2009). Although Netflix allows its users to rate the entire television series, the schema does not allow for the rating of individual episodes. The organization should examine whether their current schema supports an appropriate level of exhaustivity (Taylor, 2009). Finally, the ratings are shown in partial stars but the users are not allowed to do partial-star ratings, though this may be more of an issue of how the schema is made accessible to the user than a problem with the schema itself. Profile IssuesThe following issues illustrate the problems in the interaction between account profiles and recommendations. Netflix has added the feature to create multiple profiles within a single account for its users. This allows every user to get separately personalized recommendations. However, these profiles do not have an access to Watch Instantly. Only the main profile has access and it provides only the main profile's recommendations. Basically, a single account can have separate profiles for multiple users, but only the main profile supports all features while other profiles are restricted from accessing certain features. Netflix needs to handle better the recommendations for user profiles.Format IssuesThe Netflix schema and, in fact, the site as a whole, do not handle Blu-ray well. On the AllGenresList page, Blu-ray is listed as a genre and if it is selected, a simple list of all available Blu-rays is displayed. There is no way to filter those results into further subcategories. Also, Blu-ray is not handled well by the site search in that users are not allowed to request just Blu-ray results in the search. This is a failure to properly separate format from subject descriptors. There also exists a disconnect between the digital and physical manifestations of a single work (Lee, 2011) where Watch Instantly movies seem to work separately from DVD movies. For instance, if the user has a movie in both queues and he decides to watch it instantly, the DVD version remains in his queue. Eventually, it will show up for him even though he has already seen it which is absurd as they should be treated as the same resource.Issue and SolutionAccording to our assessment, Netflix should consider building a better ontology for TV series, franchises (including movie franchises), and other forms of entertainment that have multiple parts or formats. Currently, the service catalogs entertainment mostly by individual disc and at best by TV show season. They are unable to deliver a single page for the TV series “Lost,” for example, because they have not defined its six seasons as sub-parts of the greater whole. In the words of the FRBR model, the ontological relationships between expressions of works (for example “30 Rock”) and their manifestations in items (for example “30 Rock Season 2,” Disc 3) have not been properly articulated (Lee, 2011). Several problems arise on due to the current lack of proper ontology with respect to TV series and franchises. Some are mere annoyances: for example, when you watch the final episode of a TV show season, shouldn’t Netflix be able to take you to episode one of the following season with a single click? Right now it cannot. Another mild annoyance is the lack of a connection between Watch Instantly queue and the DVD queue. When you watch something from one queue, it will not disappear from the other; this is because Netflix’s schema focuses on the delivery method rather than the larger work, so it treats an online stream and a physical DVD of the same program as separate resources. We argue that it should not, or at least that users should be given the option to specify their preference one way or another.We noticed another ontology-related issue, this time more severe, in the way Netflix makes recommendations. For example, say the site recommends a DVD for VeggieTales, the cloyingly cute and slightly irritating children’s show. You can understand why Netflix did that, because you’ve liked other animated series, but VeggieTales is just taking things a bit too far. Try clicking the “Not Interested” button. What happened? You’ll notice that the icon below this specific Veggie Tales DVD changed to “Not Interested” but that there are several other VeggieTales recommendations waiting in line to drive you nuts. The recommendation system is a core piece of Netflix’s business, so much so that they staked a $1 million prize on it, yet they have not properly defined their ontology to allow users to exclude entire series, franchises, or other groups of related works.To solve these problems and others, we recommend that Netflix enhance their cataloging schema by defining the ontological relationships between multipart works. Their focus should not just be on the level of individual physical and digital resources, but also on more accurately defining how the various parts of a series or franchise fit together.We offer Netflix this guide on how to create a proper ontology, following Noy & McGuinness (2000). First define the domain and scope of the ontology. This should be easy to do, since we’re not interested in abstract links or relationships. Rather, we expect the links between everything in the collection to be explicit. Each resource should have the same series or franchise title somewhere on the cover. Using “Star Trek” as an example, assemble a collection of every “Star Trek” resource you have. If, after looking around for existing “Star Trek” ontologies (some enterprising Trekkie has probably already done your work for you!) you decide to build your own, enumerate all the important terms you see. In this example expect to see “Star Trek”, “The Next Generation”, “Voyager”, “Star Trek: The Movie”, “DVD”, “Stream”, “Episode”, and so on. Next, define the classes and class hierarchy using a top-down approach. Here “Star Trek” is the top node, with children movies or TV. Under movies are the various feature films, while under TV are the various series, with seasons under the series, and episodes under the seasons. Notice the difference here: where before Netflix’s schema put “discs” under seasons, we are now departing from delivery format and just focusing on the natural units provided by the Star Trek creators themselves: the episode. Instead, we will account for “DVD” and “Stream” from our terms list above as what Noy & McGuinness call “facets” in a “slot” (delivery format) that describe the episode. Finally, create your “instances”: examine each resource in turn and describe it according to your new ontology: by franchise, movie/tv, series, season, and delivery format.New rules for would allow its customers to realize the benefits of this system. Netflix will able to create a new page template for entire franchises, so “Star Trek” fans can see everything collected in one place. At the same time, people who hate “Star Trek” will be able to dismiss it from their recommendations queue in a single click. New rules could allow users to set a preference that they be served the various physical DVDs in show order, without adding eight discs to their queue individually. They will also have the option to automatically remove items from their DVD queue if those items have already been watched online, or vice-versa, since Netflix now knows to treat those episodes as equivalent (just with differing facets on the same delivery format slot).Results and ConclusionThrough ontology, Netflix would capture the underlying relationships among the concepts within the video entertainment it encompasses. Rather than requiring the user to filter through search results, “terms and definitions defined in the vocabulary can support computation and inference without a human brain pulling the levers” (Stewart, 2008, p. 163). A search for “Spock character” wouldn’t result in the top five search results being the movie “Character”, the actress Sissy Spacek, the actor Morgan Spurlock, or the movies “Charade” and “Honesty, Responsibility, and Integrity”. Rather, a defined ontology would allow Netflix to “know” that Spock is a character in the entity that is “Star Trek” and provide a range of results centered on the “Star Trek” character Spock. In essence, ontology is the path for Netflix to become “self-aware” about the realm of video entertainment. Implementation of a complete ontology would take time; however, it could be rolled out in phases. Because development of the ontology affects metadata, Netflix’s current delivery of content would not be delayed or disturbed. The initial stage would develop the classes “franchise, movie/tv, series, season, and delivery format” for Netflix’s current content. The underlying framework for this is in place, as seen by Netflix’s ability to put the “Play” or “Add” side-by-side for content that exists in both DVD and streaming. The next phase would develop further children within the content, for example characters, actors, or time period. A method could be developed where the users provide this content, similar to their ability to rate movies. With modifications to the ontology, the ultimate beneficiaries are Netflix users. As mentioned previously, implementing even a simple ontology would immediately enhance the ability to filter content from viewers or provide the option to remove items from a DVD queue once the entity has been viewed via “Watch Instantly”. Similarly, because the ontology would separate the format from the “aboutness” (Taylor, 2009) of the content, Blu-ray could be seamlessly integrated. Furthermore, by creating Netflix’s understanding of the relationships and concepts within movies and televisions, an increase in precision can be achieved when using Netflix’s search results. Search results would be enhanced with the underlying relationships and content powering the search. A search for “Naval Aviation Movie” would provide “Top Gun” and not actor “Aaron Schneider” as the top search result. However, as the ontology continues to be developed, Netflix users would not be the only beneficiary. Initially, ontology would allow Netflix content providers to more easily create focused site templates centered on particular franchises. However, as the relationships are more thoroughly developed and maintained ontology, Netflix content providers would be able provide more enhanced recommendations for its users. Rather than recommendations based on genre, content providers could query user ratings and find patterns in the content that a user rated highly. For example, a user may rate “Aliens”, “Terminator 2”, and “The King and I” highly. While on the surface no relationship can be found among these movies, ontology would allow Netflix to find patterns in these movies and recommend entertainment with a “strong female lead”. In essence, ontology provides the framework for finding what users want even when the users don’t know what they want. In order to evaluate the impact of the ontology, a few metrics could be used. First, the difference in views for “page 2, page 3, etc.” of search results could be measured. In theory, the ontology would allow more precise search results, decreasing the need to continue to click on “page 2, page 3, etc”. Additionally, the total number of DVD “rentals” and instant views should be measured before and after implementation of the ontology. Again, the ontology’s goal is to provide content relevant to the user’s needs; for Netflix, content views is the finality in user needs. Lastly, if templates based on franchises are developed, the “add” or “views” within these templates can be used another metric for proof of the ontology’s success. Ultimately, we are certain Netflix’s investment in developing ontology is the next phase for more enhanced content delivery. In a thorough evaluation of Netflix current state, there were multiple areas where Netflix could improve its taxonomy, controlled vocabulary, and use of authority control. However, we identified the development of ontology to provide the largest benefit. By properly articulating expressions of work, their manifestations, and the properties among the concepts within them, Netflix would move towards self-awareness for its own content. Ontology would allow Netflix to know its user patterns better than users themselves. More importantly, it’s the future of the internet. Netflix succeeded because they saw the future in providing streaming content before the competition (Schuster, 2010). Now, just as the internet moves towards a semantic web, Netflix must develop with the internet and begin enhancing and future proofing its search mechanism before another content provider uses a better ontology to match their users with their needs, leaving Netflix as the Blockbuster of streaming entertainment delivery. ReferencesLee, J.H. (2011). Week 1-2: Information Objects. Retrieved from University of Washington, IMT 530 Website: , N.F., & McGuinness, D.L. (2000). Ontology Development 101: A Guide to Creating Your First Ontology. Retrieved from Stanford University: , M. (2010). How Netflix Succeeded Where Blockbuster Failed. Retrieved from Yahoo! Finance: , D. (2005). “Information Retrieval” in Human-Computer Interaction Encyclopedia. , Darin L. (2008) Building Enterprise Taxonomies, (1st ed.). Portland, OR.: Mokita Press, 2008.Taylor, A. G., & Jourdrey, D. N. (2009). “Subject Analysis” In The Organization of Information. 3rd ed. Libraries Unlimited. 303-332. ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download