SIGCHI Lifetime Research Award Talk: Interdisciplinary Perspectives on Search (CHI 2020)
471
17.4
Microsoft Research336 тыс
Следующее
Популярные
49 дней – 6652:27
17.02.23 – 2 6191:01:27Опубликовано 11 июня 2021, 15:40
CHI '21: The 2021 ACM CHI Conference on Human Factors in Computing Systems
Session: Awards: SIGCHI 2020 Award Talks
Speaker: Susan Dumais
I have long been interested in how people seek information from external sources and make sense of the results. While the sources of information have continued to evolve over the years from libraries, to web search engines, to virtual assistants, many important challenges and opportunities remain. The success of any information retrieval systems depends critically on both the ability to support people in articulating their information needs and making sense of the results to solve the problem that motivated their search in the first place, as well as the need to efficiently and effectively find relevant information. My research combines these two dimensions into an interdisciplinary, user-centered perspective on information systems.
My interest in information retrieval started in the early 1980's with the observation that different people use a surprisingly wide variety of words to describe the same object or concept. This fundamental characteristic of human language set limits on how well simple word-matching techniques can do in satisfying information needs. In a paper at the pre-CHI Gaithersburg conference in 1982 [6] we describe this problem as statistical semantics. (It is symptomatic of the problem that we subsequently used vocabulary mismatch, verbal disagreement, and statistical semantics to refer to the same problem.) Over the next decade, with colleagues at Bell Labs, I developed and evaluated solutions that involved collecting multiple descriptors for objects, and reducing the dimensionality of the representation using techniques like Latent Semantic Indexing (LSI) [3][7] to mitigate the disagreement between the vocabulary that authors use in writing and searchers use to express their information needs. Similar approaches (combined with a lot more data and compute) are used to power modern word-embedding techniques that widely used in natural language processing.
Pre-recorded Presentations for CHI '21: The 2021 ACM CHI Conference on Human Factors in Computing Systems, May 8-13, 2021
Publication: microsoft.com/en-us/research/p...
Event Website: microsoft.com/en-us/research/e...
Session: Awards: SIGCHI 2020 Award Talks
Speaker: Susan Dumais
I have long been interested in how people seek information from external sources and make sense of the results. While the sources of information have continued to evolve over the years from libraries, to web search engines, to virtual assistants, many important challenges and opportunities remain. The success of any information retrieval systems depends critically on both the ability to support people in articulating their information needs and making sense of the results to solve the problem that motivated their search in the first place, as well as the need to efficiently and effectively find relevant information. My research combines these two dimensions into an interdisciplinary, user-centered perspective on information systems.
My interest in information retrieval started in the early 1980's with the observation that different people use a surprisingly wide variety of words to describe the same object or concept. This fundamental characteristic of human language set limits on how well simple word-matching techniques can do in satisfying information needs. In a paper at the pre-CHI Gaithersburg conference in 1982 [6] we describe this problem as statistical semantics. (It is symptomatic of the problem that we subsequently used vocabulary mismatch, verbal disagreement, and statistical semantics to refer to the same problem.) Over the next decade, with colleagues at Bell Labs, I developed and evaluated solutions that involved collecting multiple descriptors for objects, and reducing the dimensionality of the representation using techniques like Latent Semantic Indexing (LSI) [3][7] to mitigate the disagreement between the vocabulary that authors use in writing and searchers use to express their information needs. Similar approaches (combined with a lot more data and compute) are used to power modern word-embedding techniques that widely used in natural language processing.
Pre-recorded Presentations for CHI '21: The 2021 ACM CHI Conference on Human Factors in Computing Systems, May 8-13, 2021
Publication: microsoft.com/en-us/research/p...
Event Website: microsoft.com/en-us/research/e...
Свежие видео
12 дней – 581:13:03
13 дней – 4800:17
14 дней – 3410:26
15 дней – 129 20214:42
15 дней – 1 397 9730:58
17 дней – 037:35Случайные видео
30 дней – 51 87611:22
46 дней – 1 8060:43
120 дней – 10 5360:51
04.02.23 – 82 7129:33
03.03.22 – 1 8207:52
18.05.21 – 1 1722:27
7 дней – 7 7021:17
12 дней – 53029:40
12 дней – 1980:24
14 дней – 1 5910:45
14 дней – 2025:20
19 дней – 2 0600:58
19 дней – 1766:10
