=Paper= {{Paper |id=Vol-210/paper-22 |storemode=property |title=Reasoning with Temporal Context in News Analysis |pdfUrl=https://ceur-ws.org/Vol-210/paper22.pdf |volume=Vol-210 |dblpUrl=https://dblp.org/rec/conf/ecai/MozeticB06 }} ==Reasoning with Temporal Context in News Analysis== https://ceur-ws.org/Vol-210/paper22.pdf

Reasoning with temporal context in news analysis
Igor Mozetič and Damjan Bojadžiev1

Abstract. One aspect of implicit, contextual information is its tem- In our news analysis system, we are mainly concerned with the
poral component. Explicating this component in a formal model temporal aspects of context. The system will take into account the
makes it possible to disambiguate some context-dependent expres- time-stamp part of the metadata about news items, and temporal
sions and discover connections between expressions. We have im- models of the events reported, to distinguish related news items from
plemented and extended Allen’s algebra of temporal intervals in a unrelated ones. Thus, our working hypothesis about temporal context
reasoner that takes into account the linear nature of time and the could be expressed in the equation
granularity of temporal expressions (days/weeks/...). If this algebra
is used to model the temporal extension of events, the reasoner can temporal context(news) = temporal model + metadata(news)
track and connect the reference of indexical expressions about them.
We intend to use the reasoner for analysing news streams, to help This definition was originally inspired by [11], which deals with con-
discover connections between news items. textual vocabulary acquisition (how to infer the meaning of a new
word from textual clues). It identifies two components of context:
prior knowledge (which is subject to belief revision) and co-text of
1 INTRODUCTION the word to be learned. In our case, the task is to find a semantic
This paper expresses the combination of our interests in the subjects link between news items. The context of news is prior knowledge
of context and ontologies, taken by themselves and in their connec- in the form of a temporal model, and the metadata that comes with
tions. On the more abstract, logico-philosophical side, there are the the news. We do not deal with the model revision component, and
questions of definition and significance: what is context, what is spe- restrict our system to temporal aspects. However, causal, spatial and
cific about it, and how does it, its inclusion or its omission, affect cog- other types of models and/or ontologies also represent prior knowl-
nitive, deductive and computational processes. For example, getting edge and thus fit into our definition of context. If the restrictions to
stuck into loops, for humans and for machines, might be conceived temporality and subject matter (news items) are dropped, the equa-
as a loss of context. Judging by the many definitions of context in tion above generalizes to the form
different disciplines, the notion of context is itself context-sensitive,
and it is hard to point out the specific characteristic that distinguishes context(X) = prior knowledge + co-data(X)
context from background, prior knowledge and/or the multiplicity of
implicit facts and assumptions that is simply taken for granted, un-
noticed, left out or suppressed as too obvious to mention. This is
2 TEMPORAL ALGEBRA AND ONTOLOGY
reflected in the reluctance in some important papers on context to Allen [4] proposed an interval algebra to represent relative temporal
actually define it, such as McCarthy’s [8], and in his insistence that information, such as the order of events. The representation of events
“there is no universal context”. In connection with ontologies, there by time intervals rather than points allows the expression of hierar-
is also some context-dependence in the definition of context: ontolo- chical, indefinite and incomplete information, at different levels of
gies supply context for browsing [5] (which again indicates that con- granularity. The temporal algebra uses the thirteen possible relations
text can be practically anything), but mappings between ontologies between time intervals, such as one interval starting or finishing an-
supply context too, as in C-OWL [6]. other interval, or being before or meeting another one.
On the more computational side, our interest is in ontologies of To represent indefinite and incomplete information, Allen uses dis-
time, or of the ways we refer to its passage, and in actual imple- junction to allow any subset of the basic relations to hold between
mentations of automatic reasoning about temporal information. We two time intervals. A set of temporally related events forms a net-
have implemented Allen’s axiomatization of temporal relations, used work, with edges corresponding to (possibly disjunctive) relations
eg. in DAML-Time and SUMO [10], in the constraint logic pro- between events. There are two fundamental queries one can pose
gramming system CLP(Q) [9]. We plan to use this implementation about such a network:
in automatic news (stream) analysis, for disambiguating context-
dependent reference and for news classification. The remainder of • Find the feasible relations between all pairs of events, and
this paper gives more information about the axiomatization and its • Determine the consistency of the temporal relations.
implementation, and some examples of the intended application.
More generally, we have a hunch that some of the work done at our When we came accross this algebra, we were not aware of any
Department, eg. on user profiling and on simultaneous ontologies [7], (complete) reasoner for it. Since its networks of temporal relations
can be formulated as programming context dependency, and we are express constraints on relations between intervals, we decided to
working on a convincing formulation. implement the algebra in a constraint logic programming system
CLP(Q) [9]. The implementation allows automatic reasoning about
1 Jozef Stefan Institute, Ljubljana, Slovenia, email: igor.mozetic@ijs.si
temporal events, such as:
• “If X precedes Y, and Y overlaps with Z, what are the possible
Day1
temporal relations between X and Z ?”
• “If X takes longer then Y, can X occur during Y ?” during during
• “Given a set of temporally related events, what are the possible
Waves Earthquake
consistent scenarios on the time line ?”

On top of this basic implementation, we formulated a generic on- Figure 2. Temporal relations between both news events.
tology of time which covers everyday concepts such as hours, days,
seasons, and the relations between them. Note that there is no fixed 4) Reasoning with the tsunami model then shows that the news are
underlying time scale. This time ontology is similar to the specifica- consistent with a tsunami. Therefore, we can formulate a defeasible
tions in SUMO [3], DOLCE [2], and DAML-Time [1], and has the hypothesis: A tsunami is a possible explanation of the two events,
advantage of being executable. which links the news items in question.
5) However, if the news say that
3 NEWS ANALYSIS Waves before Earthquake
Let us first illustrate the desired feature of the analysis system by a the tsunami link will be ruled out as a possible explanation of the
simple example. Suppose we receive two news items on two subse- news sequence.
quent days: In this way, the temporal model can provide a stronger measure of
semantic similarity and thus increase the quality of the news analysis
• Day1: “Giant waves hit the shore early today.” system.
• Day2: “An ocean floor earthquake was detected yesterday.”

One interesting question that a news analyst might then ask is: Are 4 CONCLUSION
these news items related? Our definition of temporal context seems useful, especially for a
There are various techniques used for news analysis, but all essen- news analysis system, because it encompasses both static prior
tially measure the degree of similarity between items. The metrics knowledge and dynamic metadata (a sophisticated example of the use
used can be purely syntactic or increasingly based on semantics. We of such data in reasoning with abductive constraint logic programs
might roughly distinguish three levels of (semantic) similarity: is presented in [12]). If experiments with the news analysis system,
1. purely lexical, based only on the presence of keywords augmented by the temporal ontology, the constraint logic program
2. weak or lexicographic, taking into account taxonomic meaning and temporal models such as the tsunami model, prove successful,
3. strong, using models of word referents other semantic models, such as causal and spatial, will be included
too. In the tsunami example, these would be needed to capture other
The models in question are formulated in terms of the temporal on- relevant relations, such as the fact that the earthquake needs to take
tology; in the case of the news items above, a relevant example would place under the see in roughly the same geographic area.
be the temporal model of a tsunami, shown in Figure 1.
ACKNOWLEDGEMENTS
1 hour < duration(Tsunami) < 1 day
Supported by the Slovenian Research Agency, and the EU Projects
Tsunami SEKT (IST-2003-506826) and NeOn (IST-2004-27595).
starts finishes
REFERENCES
before
Earthquake Waves [1] Daml-time. http://www.cs.rochester.edu/ferguson/daml/.
or meets
[2] Dolce. http://www.loa-cnr.it/DOLCE.html.
Figure 1. A simple temporal model of a tsunami. [3] Sumo. http://www.ontologyportal.org/.
[4] J.F. Allen, ‘Maintaining knowledge about temporal intervals’, Commu-
To detect whether the news items above are related or not, we nications of the ACM, 26, 832–843, (1983).
would use the following algorithm: [5] N. Aussenac-Gilles and J. Mothe, ‘Ontologies as background knowl-
edge to explore document collections’, in Proc. RIAO-2004, pp. 129–
1) When the first news (waves) arrive, find the temporal terms 142, Avignon, (2004).
(“today”) and resolve them locally, with respect to the news meta- [6] P. Bouquet, F. Giunchiglia, F. van Harmelen, L. Serafini, and H. Stuck-
data (time-stamp). The implicit temporal reference can then be trans- enschmidt, ‘C-owl: Contextualizing ontologies’, in Proc. 2nd ISWC,
formed into explicit reference in terms of the temporal ontology, re- pp. 164–179, (2003).
[7] Grobelnik M. Fortuna, B. and D. Mladenič, ‘Background knowledge
sulting in the temporal relation
for ontology construction’, in Proc. Contexts and Ontologies, (2006).
Waves during Day1 [8] J. McCarthy, ‘Notes on formalizing context’, in Proc. 13th IJCAI,
(1993). http://www-formal.stanford.edu/jmc/context3/context3.html.
2) When the next news arrive (earthquake), the procedure gives [9] I. Mozetič, ‘Temporal interval reasoning with clp(q)’, in Proc. 8th IS-
2005, pp. 178–181, Ljubljana, (2005).
Earthquake during previous(Day2)
[10] I. Niles and A. Pease, ‘Towards a standard upper ontology’, in Proc.
2nd FOIS-2001, Ogunquit, Maine, (2001).
Here, the reference “yesterday” is expressed by applying the [11] W.J. Rapaport, ‘What is the ‘context’ for contextual vocabulary acqui-
function “previous” to the current Day2. sition?’, in Proc. 4th ICCS/ASCS-2003, pp. 547–552, Sydney.
[12] Madnick S. Zhu, Hongwei and M. Siegel, ‘Reasoning about temporal
3) Reasoning with the temporal ontology gives (Figure 2): context using ontology and abductive constraint logic programming’,
Technical Report 15, CISL, MIT, Cambridge, (2004).
previous(Day2) equals Day1