=Paper=
{{Paper
|id=Vol-1581/paper4
|storemode=property
|title=Enabled Generalized Vector Space Model to Improve Document Retrieval
|pdfUrl=https://ceur-ws.org/Vol-1581/paper4.pdf
|volume=Vol-1581
|authors=Jörg Waitelonis,Claudia Exeler,Harald Sack
|dblpUrl=https://dblp.org/rec/conf/semweb/WaitelonisES15
}}
==Enabled Generalized Vector Space Model to Improve Document Retrieval==
https://ceur-ws.org/Vol-1581/paper4.pdf
Linked Data Enabled Generalized Vector Space
Model To Improve Document Retrieval
Jörg Waitelonis, Claudia Exeler, and Harald Sack
Hasso-Plattner-Institute for IT-Systems Engineering, Prof.-Dr.-Helmert Str. 2-3,
14482 Potsdam, Germany
joerg.waitelonis@hpi.de, claudia.exeler@student.hpi.de,
harald.sack@hpi.de
Abstract. This paper presents two approaches to semantic search by
incorporating Linked Data annotations of documents into a Generalized
Vector Space Model. One model exploits taxonomic relationships among
entities in documents and queries, while the other model computes term
weights based on semantic relationships within a document. We publish
an evaluation dataset with annotated documents and queries as well as
user-rated relevance assessments. The evaluation on this dataset shows
significant improvements of both models over traditional keyword based
search.
1 Introduction
Due to the increasing demands of information seekers, retrieving exactly the
right information from document collections is still a challenge. Search engines
try to overcome the drawbacks of traditional keyword based search systems,
such as ambiguities of natural language (vocabulary mismatch), by the use of
knowledge bases, as e. g. Google’s Knowledge Graph1 . These knowledge bases
enable the augmentation of search results with structured semantic information
gathered from various sources. With these enhancements, users can more easily
explore the result space and satisfy their information needs without having to
navigate to other web sites [22, 16]. Linked Open Data (LOD) provides numerous
structured knowledge bases, but there are even more ways in which LOD can
improve search.
Many information needs go beyond the retrieval of facts. Full documents
with comprehensive textual explanations have much greater power to provide
an actual understanding than any structured information will ever have. On
the web, there are relevant documents for almost any imaginable topic. Thus, to
find the right documents is a matter of accurately specifying the query keywords.
Typically, users start with a general query and refine it when the search results do
not contain the expected result. For most cases this process works fine, because
any query string matches at least some relevant documents. The challenge of web
1
http://googleblog.blogspot.co.uk/2012/05/introducing-knowledge-graph-
things-not.html
search is thus to determine the highest quality results among a set of matching
documents. In contrast to web search, query refinement can quickly lead to empty
result sets in document collections of limited size, such as blogs, multimedia
archives, or libraries, because the wrong choice of keywords may eliminate the
only relevant document. One approach to cope with these shortcomings is to
explicitly map the document content to entities of a formal knowledge base, and
to exploit this information by taking into account semantic similarity as well as
relatedness of documents and queries.
For this purpose, we have developed and comprehensively evaluated a Se-
mantic Search system, which combines traditional keyword based search with
LOD knowledge bases, in particular DBpedia2 . Our approach shows that re-
trieval performance on less than web-scale search systems can be improved by
the exploitation of graph information from LOD resources. The main contribu-
tions of this paper are two novel approaches to exploit LOD knowledge bases in
order to improve document search and retrieval based on:
1. the adaption of the generalized vector space model (GVSM) with taxonomic
relationships,
2. measuring the level of connectedness of entities within documents instead of
traditional term frequency weighting,
Furthermore, we have published a manually assembled and carefully verified
evaluation data set with semantically annotated documents, search queries, as
well as relevance assessments at different relatedness levels3 .
In the remainder of this paper, Section 2 provides a technical instruction and
references related work. Section 3 describes the two proposed approaches and
Section 4 addresses their evaluation. The final section summarizes and discusses
the achieved results, and gives a brief outlook on future work.
2 Preliminaries and Related Work
Semantic Search makes use of explicit semantics to solve core search tasks, i. e.
interpreting queries and data, matching query intent with data, and ranking
search results according to their relevance for the query. In modern semantic
retrieval systems the ranking also makes use of underlying knowledge bases to
obtain the degree of semantic similarity between documents and queries [6]. One
of the most popular retrieval models to determine similarity between documents
and queries is the Vector Space Model (VSM). Basically, it assumes pairwise
orthogonality among the vectors representing the index terms, which means
that index terms are independent of each other. Hence, VSM does not take into
account that two index terms can be semantically related. Therefore, Wong et. al.
have introduced the Generalized Vector Space Model (GVSM), where the index
terms are composed of smaller elements and term vectors are not considered
pairwise orthogonal in general [23]. The similarity function which determines
2
http://dbpedia.org/
3
The ground truth dataset is published at: http://s16a.org/node/14
the similarity among documents dk and the query q is extended with a term
correlation ti · tj :
Pn Pn
j=1 i=1 dk,j qj ti tj
simcos (dk , q) = P
q , (1)
n
pPn
2 2
i=1 dk,i i=1 qi
where dk,i , qi represent the weights in the document and query vectors, n the
dimension of the new vectors. The term correlation ti · tj can be implemented
in different ways. Wong et. al. have used co-occurrences [23] and the model in
[17] uses WordNet, a large lexical database of words grouped into sets of syn-
onyms (synsets), each expressing a distinct concept [11]. Our approach instead
utilizes LOD resources and their underlying ontologies to determine a correlation
between related index terms.
Before exploiting the semantic relatedness, the document contents must be
annotated via Named Entity Linking (NEL). NEL is the task of identifying men-
tions of named entities in a text and linking them to the corresponding entities
in a knowledge base. A wide range of approaches for NEL exists and most of
them integrate natural language processing, such as named entity recognition,
co-reference resolution, and word sense disambiguation (WSD) with statistical,
graph-based, and machine learning techniques [19].
Our retrieval approach is inspired by the idea of Concept-based document re-
trieval, which uses WSD to substitute ambiguous words with their intended un-
ambiguous concepts and then applies traditional IR methods [5]. Several knowl-
edge bases have been exploited to define concepts. One of the first concept-based
IR approaches in [21] uses the WordNet taxonomy, whereas the more recent Ex-
plicit Semantic Analysis [3, 4] is based on concepts that have been automatically
extracted from Wikipedia.
Some approaches have already attempted to include semantic relationships
in a retrieval model. Lexical relationships on natural language words have been
applied by [7] for query expansion and by [17] in a GVSM. However, their lack of
disambiguation introduces a high risk for misinterpretation and errors. The latter
model nevertheless shows small improvements, but is limited by the knowledge
represented in WordNet. The fact that named entities play an important role in
many search queries has been considered by [1, 12]. Their approach focuses on
correctly interpreting and annotating of the query and extending the query with
names of instances of found classes.
Another approach is applied by [20, 2]. They use formal SPARQL4 queries
to identify entities relevant to the user’s information need and then retrieve
documents annotated with these entities. Since this requires knowledge of a
formal query language, it is not suited for the ordinary user.
None of the above approaches provides an allround service for end-user cen-
tered semantic search, which simultaneously builds on a theoretically sound re-
trieval model and is proven to be practically useful. Neither do any of them take
advantage of the relationships of concepts represented in a document. These
4
http://www.w3.org/TR/sparql11-overview/
are also the main points that the approaches presented in the following sections
address.
3 Linked Data Enabled GVSM
With the goal to increase search recall, the taxonomic approach uses taxonomic
relationship within the knowledge base to determine documents containing en-
tities that are not explicitly mentioned in, but strongly related to the query. We
go beyond any of the previous GVSM approaches by exploiting the semantic
relationships to also identify documents that are not directly relevant, but re-
lated to the search query. Related documents serve as helpful recommendations
if none or only few directly relevant documents exist, which is a frequent sce-
nario when searching in limited document collections. Furthermore, taxonomies
provide subclass relationships necessary for effectively answering class queries,
a special kind of topical searches, where any members of a class are considered
relevant. For example, the class search query “Tennis Players” should also re-
turn documents about instances of the class “Tennis Players”, such as “Andrew
Agassi”, “Steff Graf”, etc., even if the term or entity “Tennis Players” does not
explicitly occur in these documents.
In addition to the taxonomic approach, we also propose a connectedness ap-
proach, which aims at increasing search precision. In general, this approach com-
putes an improved term weighting by analyzing semantic relationships between
the entities within a document.
Query
Named Entity
Linking 2 1
manually verified Textual
Taxonomic Preprocessing
3a Enrichment Stemming, Stopwords, etc.
Textual 1
Documents Preprocessing
Stemming, Stopwords, etc.
Traditional Ranking A
Index Retrieval Model
2 Connectedness
Scoring 3b Connectedness 4b Ranking B
Named Entity Index Retrieval Model
Linking
manually verified
Taxonomic
Enrichment Taxonomic 4a Ranking C
3a Index Retrieval Model
Fig. 1. Evaluation architecture overview.
Fig. 1 shows the entire workflow of both proposed semantic search approaches
in addition to traditional keyword based search. The workflow consists of four
processing steps: (1) traditional syntactic document and query processing, (2)
semantic document and query annotation with LOD resources, (3) annota-
tion enrichment with semantic information, and (4) query to document match-
ing and result ranking. These steps have been implemented into the Apache
SOLR/Lucene5 indexing and retrieval system. Their purpose, theoretical back-
ground and implementation are explained in detail in the following.
The process starts with textual preprocessing (step 1), which applies stop-
word removal, basic synonym expansion, and word stemming to the document
texts. The resulting textual index terms constitute one part of the index for
both proposed approaches, so that textual and semantic index terms are treated
equally. In parallel, step 2 performs semantic document annotation by NEL with
DBpedia entities. Because NEL does not always guarantee correct results, the
annotation of documents has been manually revised with [16].
3.1 Taxonomic Enrichment & Retrieval
The taxonomic approach is a variation of the GVSM. In a GVSM, term vectors
are not necessarily orthogonal to reflect the notion that some entities or words
are more closely related or similar to each other. This section describes the
construction of term vectors from a taxonomy (step 3a) and the derived retrieval
model for matching documents to a query (step 4a).
For index term associated with an entity, the term vectors ti are constructed
from the entity vector ei of the entity it represents and the set of its classes
c(ei ):
vi X
ti = αe ei + αc , with vi = w(cj , ei ) × cj . (2)
|vi |
cj ∈c(ei )
The ei and cj are pairwise orthogonal vectors with n dimensions, where
each dimension stands for either an entity or a class. Accordingly, n is the sum
of the number of entities and the number of classes in the corpus. Taking the
cj to be orthogonal suggests that the classes are mutually independent, which
is not given since membership in one class often implies membership in another.
This model is thus not suited to calculate similarity between classes, but it does
provide a vector space in which entities and documents can be represented in
compliance with their semantic similarities.
The factors αe and αc determine how strongly exact matches of the query
entities are favored over matches of similar entities. They are constant across
the entire collection to make sure that the similarity of the term vectors cor-
responding to two entities with the same classes is uniform. To keep ti a unit
vector, they are calculated on a single value α ∈ [0, 1]:
α 1−α
αe = p and αc = p . (3)
α2 + (1 − α)2 α2 + (1 − α)2
With higher α, documents with few occurrences of the queried entity will be
preferred over documents with many occurrences of related entities.
5
http://lucene.apache.org/
Since not every shared class means the same level of similarity between two
entities, not all classes should contribute equally strong to the similarity score.
Assigning weights w(cj , ei ) to the classes within a term vector achieves this
effect. Without them, the cosine similarity of two document (or query) vectors
solely depends on the number of classes shared by the entities they contain.
The w(cj , ei ) should express the relevance of the class cj to the entity ei . We
found that Resnik’s relatedness measure [13], i.e. wResnik = maxc0 ∈S(cj ,ei ) IC(c0 )
performed best in our evaluations (cf. Sec. 4). IC(c0 ) expresses the specificity, or
information content, of the class c0 , and can be calculated by measures like linear
depth, non-linear depth [14], or Zhou’s IC [24]. It is a valuable component for our
approach because generic classes hold less information. For example, British
Explorers is a more precise description of James Cook than the general class
Person. Other similarities that include IC have been proposed in [9], [10], and
[18].
For our implementation, we have employed the classes from YAGO6 , a large
semantic knowledge base derived from Wikipedia categories and WordNet, which
interlinks with DBpedia entities [15]. Its taxonomy, which we have extend with
rdf:type statements to include the instances, is well suited for the taxonomic
approach for two reasons. First, it is fine-grained and thereby also allows for a
fine-grained determination of similar entities. Second, since the main taxonomic
structure is based on WordNet, it has high quality and consistency. This is
advantageous when using the taxonomy tree for similarity calculations. Other
taxonomies, such as the DBpedia Ontology7 or Umbel8 , also qualify.
The text index integrates into the model by appending the traditional doc-
ument vector to the entity-based document vectors. Fig. 2 shows an example
document d and query q annotated with entities (dbp:) and classes (yago:), as
well as the corresponding document vector d and query vector q.
!⃗ #⃗
!:
ENTITIES
dbp:Moon 0.7 0
“Armstrong dbp:Neil_Armstrong, dbp:Yuri_Gagarin 0 0.7
dbp:Neil_Armstrong 0
yago:Astronaut, yago:Person landed on the
0.7
yago:Astronaut 0.5 0.3
CLASSES
moon dbp:Moon, yago:CelestialBody “ yago:Person 0.3 0.2
yago:SovietCosmonauts 0 0.6
#: yago:CelestialBody 0.7 0
armstrong 1 0
“ Yuri Gagarin dbp:Yuri_Gagarin, land 1 0
WORDS
moon 1 0
yago:SovietCosmonauts, yago:Astronaut, yuri 0 1
yago:Person “ gagarin 0 1
Fig. 2. Example document and query vectors in the taxonomic model.
6
http://www.mpi-inf.mpg.de/departments/databases-and-information-
systems/research/yago-naga/yago/
7
http://wiki.dbpedia.org/services-resources/ontology
8
http://www.umbel.org
3.2 Connectedness Approach
To determine the relevance of a term within a document, term frequency (tf )
is not always the most appropriate indicator. Good writing style avoids word
repetitions, and consequently, pronouns often replace occurrences of the actual
term. However, the remaining number of occurrences of the referred-to word
depends on the writer. In documents with annotated entities, term frequency is
especially harmful when annotations are incomplete. Such incompleteness might
result, for example, when only the first few occurrences of an entity are marked,
annotations are provided as a duplicate free set on document level, or not all
surface forms of an entity are recognized.
We therefore suggest to replace the term frequency weight by a new measure,
connectedness, which requires adaptations of indexing (step 3b in Fig. 1) and
similarity scoring (step 4b). The connectedness of an entity within a document
is a variation of the degree centrality based on the graph representation of the
underlying knowledge base [8]. It describes how strongly the entity is connected
within the subgraph of the knowledge base induced by the document (document
subgraph). This approach has the desirable side effect that wrong annotations
tend to receive lower weights due to the lack of connections to other entities in
the text.
The document subgraph D, as illustrated by the example in Fig. 3 (A),
includes all entities that are linked within the document (e1 to e8 ) as well as
all entities from the knowledge base that connect at least two entities from the
document (e9 to e10 ). As connectedness is defined on undirected graphs, the
function rel(ei , ej ) is applied to to create an undirected document subgraph. It
returns true if and only if there exists some relation from ei to ej or from ej to
ei . Each entity ei ∈ D has a set Ei of directly connected entities and a set Fi of
indirectly connected entities:
Ei = {e ∈ D|rel(e, ei )} and Fi = {e ∈ D|∃x : rel(e, x) ∧ rel(x, ei )} (4)
Fig. 3 (B) illustrates Ei and Fi for the example document in part A.
A:- _________________________________ B: _________________________________
e1* e1
_________________________________
e2
e2*
_________________________________
e4
e4*
_________________________________ _________________________________
e10* _________________________________- _________________________________
e *
_________________________________
3
e3
_________________________________
e5* e9* e5
_________________________________- _________________________________
_________________________________ _________________________________
_________________________________- _________________________________
e8* e7* e6*
______________________________-
e8 e7 e6
______________________________
Fig. 3. A: Subgraph of a knowledge base spanned by a document. An arrow from entity
ei to ej indidcates that an RDF triple hei i rel hej i exists in the underlying knowledge
base. B: Each entity is connected to the members of its Ei by a solid line, and to the
members of its Fi by a dashed line.
Based on these sets, connectedness cn(ei , d) is computed as follows:
|D| X
cn(ei , d) = 1 + (|Ei | + |Fi |) × , where nd = |Ej | + |Fj |. (5)
nd
ej ∈D
Entities may have no connections to any other entities in the document sub-
graph. Since they are nevertheless relevant to the document, we add 1 to all
scores. The multiplication with |D|/nd normalizes the score by the average num-
ber of connected entities over all e ∈ D. This normalization creates comparability
between different documents, which is otherwise lacking for two reasons: On the
one hand, entities are more likely to be connected to other entities in documents
with more annotations. On the other hand, a single connection to another entity
is more significant in a sparse document subgraph than in a dense one. The aver-
age score is preferable over the sum of scores because it keeps a larger variation
in scores.
While calculating the term’s weight within a document via connectedness,
we keep the traditional inverse document frequency (idf ) to calculate the term’s
distinctness. Whether or not a word or entity has a large power to distinguish
relevant from non-relevant documents depends on the document corpus. In a
corpus with articles about Nobel Prize winners, for example, ’Nobel Prize’ is
a common term, whereas in general collections, the same term is much less
frequent, and its occurrence is more informative. Distinctness can thus only be
accurately estimated by a corpus dependent measure.
Since connectedness is independent of taxonomic classes, for this approach
the term vectors consist of the entity vector (all 0 for unannotated terms) con-
catenated with the traditional term vector. While weights in the traditional term
vector part remain tf-idf weights, the entity vectors’ values are cn-idf values, i.e.
|D| |N |
w(ei , d) = cn(ei , d) × idf (ei ) = 1 + (|Ei | + |Fi |) × × log . (6)
nd df (t)
On the other hand, connectedness is not suitable for weighting query enti-
ties. The main reason for this is that most of the times queries are too short
to contain sufficiently many connections to convey any meaningful context. Fur-
thermore, whether query weights are recommendable is application-dependent,
and consequently left out in our considerations.
4 Evaluation
The evaluation shows how the proposed retrieval models improve the search
effectiveness. To perform an initial optimization, we have manually assembled a
ground truth dataset with documents, queries, and relevance judgements. Since
human relevance judgements are idiosyncratic, variable and subjective, we have
subsequently conducted a multi-user study with different judges, to double check
whether the proposed method also performs well in a real user scenario.
An appropriate evaluation of the presented methods necessitates a correctly
annotated dataset. This prohibits the use of traditional retrieval datasets, such
as large scale web-search datasets, e. g. as provided by the TREC9 community,
because the semi-automatic creation of necessary semantic annotations would
have taken too long. Datasets for NEL evaluation are perfectly annotated, but
do not provide user queries and relevance judgments [19]. Since no appropriate
dataset was publicly available, we decided to compile a new dataset. It consists
of 331 articles from the yovisto blog10 on history in science, tech, and art. The
articles have an average length of 570 words, containing 3 to 255 annotations
(average 83) and have been manually annotated with DBpedia entities [16].
Inspired by the blog’s search log, we have assembled and also manually annotated
a set of 35 queries. The blog authors, as domain experts, assisted in the creation
of relevance judgements for each query.
On this initial dataset, we optimized some parameters of our approaches
with respect to Mean Average Precision (MAP) and Normalized Discounted
Cumulative Gain (NDCG). The Resnik similarity with Zhou’s IC performed
best for the class-based method, and connectedness worked best when average
normalization (as described in Section 3.2) was applied. For text-only search
on this dataset, length normalization was not beneficial, and the classical linear
term frequency weighting performed better than Lucene’s default, which takes
the square root of the term frequency.
With this configuration, we have set up a user study. Since ranking all 331
documents for 35 queries was beyond our capacities, we have used pooling to
identify potentially relevant documents. For every query, the top 10 ranked doc-
uments from text-, class- (Resnik-Zhou), and connectedness search have been
collected and presented to the users in random order. The users were then asked
to assign every document to one of the following five categories based on its rela-
tion to the query: Document is relevant (5), parts are relevant (3), document is
related (3), parts are related (1), irrelevant (0). The rounded arithmetic mean of
the relevance scores (indicated in parentheses) determines all relevance score in
the ground truth. Furthermore, the participants directly compared the rankings
produced by text search (baseline), class-based search, and connectedness-based
search and identified the best (score 2) as well as the second-best ranking (score
1). In total, 64 users have participated in the relevance assessments. All queries
have been assessed by at least 8 participants.
Table 1. Evaluation Results
Method MAP NDCG MAP@10 NDCG@10 RR Prec@1
Text (baseline) 0.696 0.848 0.555 0.743 0.960 0.943
Concept+Text (α = 1) 0.736 0.872 0.573 0.761 0.979 0.971
Connectedness (only) 0.711 0.862 0.567 0.752 0.981 0.971
Connectedness (with tf) 0.749 0.874 0.583 0.766 0.979 0.943
Taxonomic (no similarity, α = 12 ) 0.766 0.875 0.603 0.758 0.961 0.943
Taxonomic (Resnik-Zhou, α = 12 ) 0.768 0.877 0.605 0.762 0.961 0.943
9
http://trec.nist.gov/
10
http://blog.yovisto.com/
Tab. 1 shows that the inclusion of semantic annotations and similarities
clearly improves retrieval performance compared to traditional text search. As
expected, the taxonomic approach increases overall recall (97.5% as compared to
93.2% for text search) because it is able to retrieve a larger number of documents.
However, it also improves the ranking, measured by MAP and NDCG.
The connectedness approach performs better than text search, but worse
than the other semantic methods, including the simple “Concept+Text” ap-
proach, where entities are treated as regular index terms within Lucene’s default
model. This poor performance is surprising because the results from the users’
direct comparison of the three rankings indicates that connectedness (with an
average score of 1.09) provides better rankings than taxonomic (1.01) and text
search (0.90). This seeming contradiction hints at a difference between informa-
tion retrieval evaluation measures and user perception of ranking quality. The
evaluators have judged mainly by the very top few documents. Connectedness
outperforms the other approaches in this respect, as shown by the reciprocal
rank (RR) and precision@1 in Tab. 1. Also, connectedness performs best when
related documents are not considered relevant.
Combining the connectedness measure with tf weights seems to be the best
weighting. When considering only the first 10 documents, as users often do, this
combined weighting’s performance can be considered slightly better than the
taxonomic approach due to its higher NDCG value, because NDCG takes differ-
ent relevance levels into account, while MAP does not. However, connectedness
is inferior to the taxonomic approach on the complete search results, because it
simply does not retrieve certain documents. This harms the recall and negatively
effects MAP and NDCG, while precision may still be higher.
Table 2. Comparison of semantic similarities for the taxonomic model
Similarity Uniform Jiang-Conrath [9] Lin [10]
Specificity – lin depth log depth Zhou’s IC lin depth log depth Zhou’s IC
MAP 0.766 0.753 0.766 0.767 0.767 0.766 0.767
NDCG 0.875 0.868 0.875 0.875 0.876 0.875 0.877
Similarity Resnik [13] Tversky Ratio [18] Tversky Contrast [18]
Specificity lin depth log depth Zhou’s IC – –
MAP 0.768 0.768 0.768 0.768 0.763
NDCG 0.877 0.876 0.877 0.876 0.873
Tab. 1 indicates that the use of the Resnik-Zhou similarity only has a very
small positive impact on retrieval compared to the same approach with uniform
weights. This is in line with the numbers from Tab. 2, which demonstrate that
the choice of semantic similarity has only little impact. Apparently, the number
of shared classes has a more significant influence in this setup than the class
weights.
To determine the influence of NEL quality, we have also executed experiments
with documents, where annotations have not been revised after automated NEL.
The results still show an improvement over text search, but a MAP is 2.5% to
4.5% and NDCG 0.1% to 1.6% lower than in the equivalent experiments on
manually revised documents.
5 Conclusions and Future Work
This paper has shown how Linked Data can be exploited to improve document
retrieval based on an adaption of the GVSM. We have introduced two novel ap-
proaches utilizing taxonomic as well as connectedness features of Linked Data re-
sources annotated within documents. Our evaluation has shown that both meth-
ods achieve a significant improvement compared to traditional text retrieval. The
connectedness approach tends to increase precision whereas the taxonomic ap-
proach raises recall. Thereby, the similarity measure that weights taxonomic
classes of index terms has only little influence on the retrieval. The quality of
annotations substantially impacts the retrieval results, but uncorrected state-of-
the-art NEL still ascertains an improvement. In general, the quality of Linked
Data itself is an obstacle. Only about 65% of entities in the dataset used have
type statements, which are essential for the performance of the taxonomy-aided
retrieval. Since no annotated ground truth datasets with relevance judgements
exist, we have created one via a pooling method.
There are still open questions, which are to be answered in future work, in-
cluding how well the models perform with other knowledge bases (e. g. Wikidata
or national authority files) or languages, what other semantic relations between
entities are valuable for document retrieval, and how the semantic similarity
can obtain more influence. The annotation of queries with classes may improve
the retrieval, and so could the combination of the two proposed methods. Fact
retrieval would benefit from the more precise connectedness approach whereas
increasing the impact of the taxonomic approach would be preferable also in
exploratory search system. Furthermore, the main ideas can be transferred to
an adapted language or probabilistic retrieval model.
References
1. T. H. Cao, K. C. Le, and V. M. Ngo. Exploring combinations of ontological features
and keywords for text retrieval. In T. B. Ho and Z.-H. Zhou, editors, PRICAI,
volume 5351 of LNCS, pages 603–613. Springer, 2008.
2. P. Castells, M. Fernandez, and D. Vallet. An adaptation of the vector-space model
for ontology-based information retrieval. IEEE Transactions on Knowledge and
Data Engineering, 19(2):261–272, Feb 2007.
3. O. Egozi, E. Gabrilovich, and S. Markovitch. Concept-based feature generation and
selection for information retrieval. In AAAI, volume 8, pages 1132–1137, 2008.
4. O. Egozi, S. Markovitch, and E. Gabrilovich. Concept-based information retrieval
using explicit semantic analysis. ACM Transactions on Information Systems, 2011.
5. F. Giunchiglia, U. Kharkevich, and I. Zaihrayeu. Concept search. In Proc. of the
ESWC 2009, pages 429–444, Berlin, Heidelberg, 2009. Springer-Verlag.
6. S. Harispe, S. Ranwez, S. Janaqi, and J. Montmain. Semantic measures for the
comparison of units of language, concepts or entities from text and knowledge base
analysis. CoRR, abs/1310.1285, 2013.
7. A. Hliaoutakis, G. Varelas, E. Voutsakis, E. G. M. Petrakis, and E. E. Milios.
Information retrieval by semantic similarity. Int. Journal on Semantic Web and
Information Systems, pages 55–73, 2006.
8. M. O. Jackson et al. Social and economic networks, volume 3. Princeton University
Press, 2008.
9. J. J. Jiang and D. W. Conrath. Semantic similarity based on corpus statistics and
lexical taxonomy. 1997.
10. D. Lin. Extracting collocations from text corpora. In First Workshop on Compu-
tational Terminology, pages 57–63, 1998.
11. G. A. Miller. Wordnet: A lexical database for english. Commun. ACM, 38(11):39–
41, Nov. 1995.
12. V. M. Ngo and T. H. Cao. Ontology-based query expansion with latently related
named entities for semantic text search. Advances in Intelligent Information and
Database Systems, 2010.
13. P. Resnik. Using information content to evaluate semantic similarity in a taxonomy.
In Proc. of the 14th Int. Joint Conference on Artificial Intelligence, pages 448–453,
San Francisco, CA, USA, 1995. Morgan Kaufmann Publishers Inc.
14. N. A. L. Seco. Computational models of similarity in lexical ontologies. Technical
report, Masters Thesis, 2005.
15. F. M. Suchanek, G. Kasneci, and G. Weikum. Yago: a core of semantic knowledge.
In Proceedings of the 16th international conference on World Wide Web, pages
697–706. ACM, 2007.
16. T. Tietz, J. Waitelonis, J. Jäger, and H. Sack. Smart media navigator: Visualiz-
ing recommendations based on linked data. In 13th International Semantic Web
Conference, Industry Track, pages 48–51, 2014.
17. G. Tsatsaronis and V. Panagiotopoulou. A generalized vector space model for text
retrieval based on semantic relatedness. In Proceedings of the 12th Conference of
the European Chapter of the Association for Computational Linguistics: Student
Research Workshop, pages 70–78. Association for Computational Linguistics, 2009.
18. A. Tversky. Features of similarity. Psychological Review, 84(4):327, 1977.
19. R. Usbeck et al. GERBIL – general entity annotation benchmark framework. In
24th WWW conference, 2015.
20. D. Vallet, M. Fernández, and P. Castells. An ontology-based information retrieval
model. The Semantic Web: Research and Applications, 2005.
21. E. M. Voorhees. Using wordnet to disambiguate word senses for text retrieval. In
Proceedings of the 16th annual international ACM SIGIR conference on Research
and development in information retrieval, pages 171–180. ACM, 1993.
22. J. Waitelonis and H. Sack. Towards exploratory video search using linked data.
Multimedia Tools and Applications, 59(2):645–672, 2012.
23. S. K. M. Wong, W. Ziarko, and P. C. N. Wong. Generalized vector spaces model
in information retrieval. In Proc. of the 8th SIGIR Conf. on Research and Devel-
opment in Information Retrieval, pages 18–25, New York, NY, USA, 1985. ACM.
24. Z. Zhou, Y. Wang, and J. Gu. A new model of information content for seman-
tic similarity in wordnet. In Future Generation Communication and Networking
Symposia, 2008. FGCNS ’08. 2nd Int. Conf. on, volume 3, pages 85–89, Dec 2008.