Common use of Needs Clause in Contracts

Needs. A key challenge is understanding what informa- tion needs users have that the system should support. What these needs are, and how they are expressed, often depend on the device as well as the data. A personal digital assistant may also address these information needs proactively, e.g., identify routing tasks and pulling up all the related material, like preparing a travel expense declaration. Task Representation, Identification & Abstraction. Once we understand something of personal search tasks, we need to represent access patterns, information needs, and behaviours in a way that existing systems can use to reason, and researchers can use to investigate new systems. This would involve extracting tasks from private data; abstracting them to allow insight; and developing a common protocol for describing tasks, or classes of tasks, without violating privacy or security. Index and Schema Representation. A similar early research challenge is aggregation and representation of heterogeneous data sources and formats. One has to come up with a represen- tation generic enough for data that ranges from unstructured to fairly structured, and sometimes with comprehensive metadata. Efforts on data integration (e.g., by the database community) and common vocabularies such as ▇▇▇▇▇▇.▇▇▇ and Dublin Core (e.g., by the semantic web community) may be helpful to this end. As a community, we could distinguish better between the logical and physical representations of information, and express retrieval models at the logical level while delegating the actual relevance estimation over heterogeneous information in various data silos to the underlying physical layer. Once the data has been represented in a common format, one has to think about suitable ways of querying it and interfaces to expose to users. While a rich query language might be helpful as an intermediary, it is unlikely to be apt for common users who would rather express their information needs using natural language. Linking and Disambiguation across Silos. Once representational issues are addressed, a higher order research challenge is extracting entities from the heterogeneous data repositories - performing disambiguation, if required, and then linking entities within the collection. A more difficult challenge is linking of entities/objects to particular tasks, i.e. finding all the relevant artifacts associated with a given task, or set of tasks. This will enable a personal knowledge graph that contextualizes the relationships in user data. One challenge here is that entities in the personal knowledge graph are unlikely to be generally popular, so that signals commonly used for named entity disambiguation (e.g., popularity and coherence) may just not work. Alternative signals (e.g., co-access patterns, similarity of usernames and email addresses) can potentially serve as replacements. Ranking and Retrieval. Challenges related to search in personal data include the heterogeneity of information access tasks (ranking, summarization, etc.), of data sources, and of type of inter- actions (depending on device and modality). Retrieval methods will have to take the different characteristics of the data into account and also make use of metadata (e.g., creation times of files). It is foreseeable that the type of query result will depend on the information need at hand and may take the form of a list of files, bundles of interlinked files, or even a summary generated from the contents of relevant files. In a new opportunity for IR, the searcher in a personal system may also be the author or curator. Their work, e.g. in filing into folders, can inform the retrieval process. Computing over Aggregate Personal Data. Personal data provides a very rich representation of an individual’s content and behavioral interaction patterns. Aggregating across individuals could augment this allowing generalization to new contexts. For example, user interaction data is an important signal for learning-to-rank models in web search; these models require observing interactions across many users for the same query-document