New Applications. In a truly interactive IR system, such as a conversational information seeking system, information needs are complex, typically requiring iterative user dialogue with the system, with each iteration encompassing query reformulation and access to the index. The costs of these iterations might be reduced in several dimensions. For example, we could leverage incremental computations across the sequence of iterations to enhance scalability and efficiency through suitable caching or prediction of the remaining dialogue the user will engage in. The development of real-time search applications on the Internet of Things (IoT) infrastructure also requires new indexing and search architectures, to allow the seamless ingestion, indexing and querying of data in real-time. Related to this, the emer- gence of new search services in the cloud, with service level guarantees coupled with limited resources and various constraints, opens up a number of unexplored directions in search efficiency and effectiveness trade-offs. For example, the search engine has to continuously analyze and decide on the best configuration of its system given the available resources and the guaranteed level of services. Moreover, the increasing importance of ensuring account- ability, transparency, and explainability in machine learned MSSs entails additional costs for the search engine, beyond the actual task of retrieving information. Such costs include the generation of the explanations for the retrieved results and their visualization, which require new efficiency research directions including the revisiting of data structures to cope with such additional costs.
Appears in 3 contracts
Sources: End User Agreement, End User Agreement, End User Agreement