Politecnico di Milano Department of Electronics, Information and Bioengineering Exploratory Computing: designing engines for discovery-driven user experiences Luigi Spagnolo [email protected]December 10, 2012 L. Spagnolo Exploratory Computing 1 / 39
45
Embed
Exploratory computing: designing discovery-driven user experiences
A presentation of my PhD thesis work on exploratory user experiences
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Politecnico di MilanoDepartment of Electronics, Information and
Bioengineering
Exploratory Computing: designing enginesfor discovery-driven user experiences
Beyond information seeking: making sense, understanding,investigating.• Example: digital library of scientific publications
Retrieval taskFinding recent, significant publications on a certain research topicand/or by certain authors
Exploration
• (E.g. scholar) acquiring knowledge about a research domain orspecific community: approaches, terminology, main sources ofcontribution, communication style, etc.
• (E.g. university reviewers): Evaluating the contribution of aperson/research group (citations, co-authors, metrics, etc.)
L. Spagnolo Exploratory Computing 3 / 39
Information exploration | Traditional approaches
• Hypertext navigationI Based on well-engineered information architectures
(hierarchies of hyperlinked pages)I Static taxonomies fail to scale to large bodies of informationI Interesting content buried under levels of navigation and
information overload• Keyword-based search
I Arranges content items dynamically according to user needs(expressed by a query)
I Not very suitable for ill-defined knowledge, non-retrieval goalsI The user is required to know the terminology
• Emerging patterns combine search and browsing →exploratory search
I Faceted navigation: iterative filtering according to multiplecriteria (facets). Becoming a de facto standard.
I Limit: focus is still mainly on object seekingI A more comprehensive approach to the problem is needed
L. Spagnolo Exploratory Computing 4 / 39
Information exploration | My proposal
FELIS: Faceted Exploration for Large Information SpacesA comprehensive, general framework for exploratory userexperiences, covering:• Knowledge representation according to multiple itemproperties (facets) and querying
• Widgets and interaction patterns for (semantic) facetedexplorations
• information visualization strategies to highlight thecorrelation between item features, supported by explorationmetrics to estimate their relevance
L. Spagnolo Exploratory Computing 5 / 39
Outline
Information exploration
The proposed model
The interface
Conclusions
Publications
L. Spagnolo Exploratory Computing 6 / 39
The proposed model | How the exploration works
• Over a structured information baseI Represented according to a more or less complex semanticsI Basic faceted classification: categories grouped by propertyI Semantic faceted classification: also the properties of
related items are used (e.g. browsing artworks by artist’scountry)
• The user iteratively defines the scope of exploration within thewhole space
I The query specifies the features that information items shouldor should not possess
• The user gets the set of items matching the query plusfeedback information about the properties shared by suchresults
• Shared features are also used to refine the query restricting(zoom-in) or enlarging (zoom-out) the scope of theexploration.
L. Spagnolo Exploratory Computing 7 / 39
The proposed model | FelisQL (1)
FelisQL: knowledge modelling and query language• An extension to LISQL (Prof. Ferré, University of Rennes I)• Item representation and querying, definition of facetedtaxonomies, syntactic query transformations
• Fully translatable into standard RDF and SPARQLWhy yet another language?!• user-centric, high level representation of concepts (forboth item description and querying) in a concise and naturalway (e.g. avoiding unnecessary variables)
• easier graphical representation into interface elements• Understandable by domain experts (designers and powerusers) with lack of expertise in programming
I Easier configuration and fast prototyping of exploratoryapplications (future work: computer-aided design tools)
L. Spagnolo Exploratory Computing 8 / 39
The proposed model | FelisQL (2)
Examples of representable queries:• Publications authored by foreign PhD students at Politecnicodi Milano, with the support of a professor from a foreignuniversitya publication and autho r :
[ a ’PhD student’ and coun t r y : not Italy ]at a f f i l i a t i o n : ’Politecnico di Milano’ ,
[ a professor ]at a f f i l i a t i o n : coun t r y : not Italy
• Italian painters authors of landscapes in XIX or XX centurya painter
and coun t r y : ’Italy’and autho r of
[ s u b j e c t : [ a landscape ]and p e r i o d :
"XIX century" or "XX century" ]
L. Spagnolo Exploratory Computing 9 / 39
The proposed model | FelisQL (3)
Example of description:’Monna Lisa’ [ a painting
and s u b j e c t : [ a woman and h a i r : c o l o r : "black" ] ,[ a landscape ]
and autho r : ’Leonardo Da Vinci’ [a painter and an engineer and a scientistand n a t i o n a l i t y : "Italian"and b i r t h : [
p l a c e : Vinci [ a city andpart of ’Province of Florence’ ]
and yea r : 1452 ]and death : [
p l a c e : Amboise [ a city andpart of ’Indre et Loire’ ]
and yea r : 1519 ]] . . .
]
L. Spagnolo Exploratory Computing 10 / 39
The proposed model | FelisQL (4)
Query transformations: syntactic operations (supported out byhyperlinks in the interface) allowing to move from a stage toanother of the exploration, e.g.:• Focus change: selecting the specific sub-expression (focus) ofthe query at which appending a new restrictiona painting and artist : [a woman]〈focus a woman〉
• And-insertion: adding a new restriction in conjunction at theselected focusa painting and artist : [a woman]〈and country : ‘UK’〉
• Or-insertion adding a new restriction in disjunction at theselected focusa painting and artist : [a woman and country : ‘UK’]〈or country : ‘USA’〉
• Exclusion: selecting items that do not have certain featuresa painting and artist : [a woman and country : ‘UK’ or ‘USA’]〈exclude subject : [a portrait]〉
L. Spagnolo Exploratory Computing 11 / 39
The proposed model | Relevance measures (1)
Providing feedback about the interestingness of a item feature Cw.r.t. the whole information space Ω or the specific context ofexploration represented by the query qNotation: | C′Ω | is the number of items in Ω matching C′.
• Relative count: most common measure
µcount(C, q) =∣∣∣q〈and C〉
Ω
∣∣∣• Frequency: the percentage of items that match the query qand also the feature C
µfreq(C, q) =µcount(C, q)
|qΩ|
I Corresponds to the confidence of the association rule q → C,i.e. conf(q → C)
L. Spagnolo Exploratory Computing 12 / 39
The proposed model | Relevance measures (2)
• Recall: the percentage of items that match the query qamong those matching the feature C
µrecall(C, q) =µcount(C, q)∣∣Cq
Ω
∣∣I Corresponds to conf(C → q)I Measures peculiarity, i.e. how much C is typical in the context
of q (and less likely to be found elsewhere)• Highlighting significant correlations between a feature Cand the query q
I estimating how much both frequency and recall diverge fromtheir expected values if C and q where statistically independent
L. Spagnolo Exploratory Computing 13 / 39
The proposed model | Relevance measures (3)
Civilization orperiods related toarchaeologicalvenues in Italy.• Font size:frequency/countof terms
• Bar charts:frequency andrecall When nofilter is set, therecall is 100%.
Civilization orperiods related toarchaeologicalvenues in NorthernItaly.• Coloured tags:significantpostive/negativefrequency (textcolor) or recall(background)
• Bar charts: Detailand deviancefrom expected.
L. Spagnolo Exploratory Computing 15 / 39
The proposed model | Relevance measures (5)
Civilization orperiods related tomuseums inSouthern Italy.
L. Spagnolo Exploratory Computing 16 / 39
The proposed model | Relevance measures (6)
Joint relevance measure: measuring how much the features Ca
and Cc are correlated in the context of the query q• E.g. Lift:
νlift(Ca, Cb, q) =
∣∣∣q〈and Ca〉〈and Cb〉
Ω
∣∣∣∣∣∣q〈and Ca〉
Ω
∣∣∣ · ∣∣∣q〈and Cb〉
Ω
∣∣∣
L. Spagnolo Exploratory Computing 17 / 39
The proposed model | Relevance measures (7)
Correlations betweenconcepts representingCivilization orperiods related toarchaeologicalvenues in CentralItaly.• Node: count(size) andsignificantfrequency (color)
• Edge size andcolour: Liftmeasure betweenpairs of concepts
L. Spagnolo Exploratory Computing 18 / 39
The proposed model | Relevance measures (8)
Correlations betweenconcepts representingCivilization orperiods related toarchaeologicalvenues in CentralItaly.• Selection of anode (filteringpreview)
L. Spagnolo Exploratory Computing 19 / 39
The proposed model | Difference metrics (1)
Providing feedback about how much the relevance measure of aitem feature C changes when moving from a new stage of thetexploration, e.g. from query qn−1 to query qn• Difference metrics can be applied to the result of any querytrasformation
• providing additional clues of possible correlations to beinvestigated
• Absolute frequency difference:
∆freq(C, qn) = µfreq(C, qn)− µfreq(C, qn−1)
• Relative frequency difference:
δfreq(C, qn) =∆freq(C, qn)
µfreq(C, qn)
L. Spagnolo Exploratory Computing 20 / 39
The proposed model | Difference metrics (2)
Bipolar bar chart• showing frequency differences for types of archaeological venue• previous query:location : ‘Southern Italy’ or ‘Insular Italy’
• current query: location : ‘Central Italy’
L. Spagnolo Exploratory Computing 21 / 39
The proposed model | Difference metrics (3)
Difference bar chart• showing frequencydifferences forperiods andcivilizationsassociated toarchaeologicalvenues
• previous query:location :
‘Southern Italy’ or‘Insular Italy’
• current query:location :
‘Central Italy’
L. Spagnolo Exploratory Computing 22 / 39
Outline
Information exploration
The proposed model
The interface
Conclusions
Publications
L. Spagnolo Exploratory Computing 23 / 39
The interface | Main elements (1)
• Facet widgetsI Represent a specific faceting property ( mono-dimensional)I Allow to apply restrictions over property values (one or more
query transformations are implemented)I Show the distribution of property values (according to one
ore more relevance metrics)I Possible visualizations depend on: type of property (functional
vs. multivalued) and communicative vs. analytical purposes• Canvases
I Provide preview and access to information itemsI Possibly provide aggregate view according to two or more facet
properties (multidimensional) at a given level of granularity• (Interactive) query representation
I Shows the current selection of filters (can be embedded infacet widgets)
I Allows for focus change in semantic faceted exploration
L. Spagnolo Exploratory Computing 24 / 39
The interface | Main elements (2)
Demo: PoliculturaPortal www.policulturaportal.it: allowsexploring 600+ interactive narratives produced within thePolicultura context for schools.
Features for this focus (58 items of type Artist ):
a painting
period: orXIX Century XX Century
author: gender: female
Others (show)Tamara de Lempicka Suzanne Valadon Natalia Goncharova Mary CassattLouise Bourgeois Gwen John Georgia O'Keeffe Frida Kahlo Evelyn De MorganElizabeth Thompson Elizabeth SiddalBerthe Morisot 21
Features for this focus (58 items of type Artist ):
a painting
period: orXIX Century XX Century
author: gender: female
Others (show)Tamara de Lempicka Suzanne Valadon Natalia Goncharova Mary CassattLouise Bourgeois Gwen John Georgia O'Keeffe Frida Kahlo Evelyn De MorganElizabeth Thompson Elizabeth SiddalBerthe Morisot 21
Features for this focus (58 items of type Artist ):
a painting
period: orXIX Century XX Century
author: gender: female
Others (show)Tamara de Lempicka Suzanne Valadon Natalia Goncharova Mary CassattLouise Bourgeois Gwen John Georgia O'Keeffe Frida Kahlo Evelyn De MorganElizabeth Thompson Elizabeth SiddalBerthe Morisot 21
Features for this focus (58 items of type Artist ):
a painting
period: orXIX Century XX Century
author: gender: female
Others (show)Tamara de Lempicka Suzanne Valadon Natalia Goncharova Mary CassattLouise Bourgeois Gwen John Georgia O'Keeffe Frida Kahlo Evelyn De MorganElizabeth Thompson Elizabeth SiddalBerthe Morisot 21
and University of Lugano research projects portal• JavaScript APIs allowing for fast prototyping of applications(both client-side only and based on Solr as search server)
I Towards computer-aided tools for design and protyping (futurework)
• Concepts for advanced semantic faceted explorationI Future work: obtaining a prototype based on Sewelis
(University of Rennes 1) as search server
L. Spagnolo Exploratory Computing 35 / 39
Outline
Information exploration
The proposed model
The interface
Conclusions
Publications
L. Spagnolo Exploratory Computing 36 / 39
Publications | So far I
JournalsSpagnolo, Luigi, Davide Bolchini, Paolo Paolini, and Nicoletta DiBlas (2010). “Beyond Findability: Search-Enhanced InformationArchitecture for Content-Intensive Rich Internet Applications”. In:Journal of Information Architecture 2.1, pp. 19–36.
Conference proceedingsDi Blas, N., P. Paolini, and L. Spagnolo (2012). “PoliCulturaPortal: 17,000 Students Tell their Stories about Cultural Heritage”.In: Museums and the Web 2012.De Caro, Stefano, Nicoletta Di Blas, and Luigi Spagnolo (2010).“In SEARCH of Novel Ways to Design Large Cultural HeritageWebsites”. In: Museum and the Web 2010 Proceedings. Ed. byJ Trant and D Bearman. Archives and Museum Informatics,Toronto.
L. Spagnolo Exploratory Computing 37 / 39
Publications | So far II
De Caro, Stefano, Nicoletta Di Blas, Paolo Paolini, andLuigi Spagnolo (2009). “Search-enhanced web informationarchitecture for findability and discovery of archeological heritage:the MiBAC-DGA case study”. In: Congresso Nazionale AICA.
L. Spagnolo Exploratory Computing 38 / 39
Publications | In progress
• L.Spagnolo , P. Paolini, and F. Ferré“Exploratory computing: a new paradigm for accessing largebodies of information”ACM Trans. on Information Systems
• L.Spagnolo and S. Ferré “A flexible, user-centric language forsupporting faceted explorations”IEEE Trans. On Knowledge and Data Engineering
• L.Spagnolo and P. Paolini“Advanced widgets for faceted visual exploration”ACM Trans. On Computer-Human Interaction