Query by Example: Concepts

Once the Shared Perspectives are available for every item, if a user wanted to execute a SQL query on them, he would need to know exactly the perceptual features available for query, as well as the values they are looking for. Given that extraction methods can result in any number of perceptual tuples, some of them with implicit features, such SQL query is not feasible. A similar situation exists in multimedia databases, where items are also represented with tuples of low level features. Instead of a SQL query, the user can provide an item with a high-level representation that resemble what they are looking for in a QBE: they can give the system an image of a tree on a hill to get a visually similar one back. We propose to use this approach for experience items, where the user can provide a movie with a nice story that follows the book, and get similar movies based on that perception.

The use of QBE not only avoids the problem of unfeasible queries on implicit features, but also the challenge of having implicit target, where users don't know what they're looking for. If they knew what they were looking for, a SQL query would suffice. In addition, QBE can be an iterative process, where the user can select an item from the results as the new example, until an item is found with the desired perceptual features. Conceptually, the process of Query by Example with Shared Perspectives is described below, using Shared Perspectives Concepts.

• User U is thinking of a certain Experience Item: i₀ that fits their current situation or mood, and they want something that can provide a similar experience.
• The user provides i₀ as the example of the QBE. This item has n Shared Perspectives, therefore SPT₀:{SPT₀₀,SPT₀₁,...,SPT_0n}. The system contains the SPTs of each of the k experience items: i₀, i₁, ..., i_k in its collection C.
• For each SPT in SPT₀, the system will calculate the similarity with the rest of the SPT: SPT₁, SPT₂,...,SPT_k available in the table SPR.
• The system will then select m most similar SPTs for each one of SPT₀.
• Then it will show the user a display d₀ with them items that have the most similar SPT for each SPT in SPT₀. Therefore d₀ will have a size n x m.
• If any of those items is satisfactory for U, then the process is over.
• If not, the user can select the the item they like the most from d₀ and provide this as the new example of QBE.
• This process is iterative, it can be executed until an item is found, however, the SPTs of items shown in d₀ are discarded when computing d₁, then the ones in d₀ and d₁ for d₂, etc.