This paper reports on the results of our experiments in the Monolingual English, German and Portuguese tasks and the Bilingual German topics on English collections, English topics on German collections and English topics on Portuguese collections tasks. Seven runs were submitted as o cial runs, four for the monolingual task and three for the bilingual task. We used the Terrier (TERabyte RetrIEveR) Information Retrieval Platform version 2.1 to index and query the collections. Experiments were performed for both tasks using the Inverse Document Frequency model with Laplace after-e ect and normalization 2. Topics were processed automatically and the only elds considered were the title and the description. We included the title eld only for an experiment with the Portuguese collection. The stopword list provided by Terrier was used to index all the collections. Results for both the monolingual and bilingual tasks were low in terms of precision and recall mainly due to the following reasons: 1) no manual processing was done; 2) no query expansion based on automated relevance feedback was added; 3) no experiments including the narrative eld were run; 4) no terms were translated for the bilingual task; 5) no German and Portuguese stopword lists were used instead of the default stopword list; and 6) no pre-processing or removal of diacritic marks was performed. We are running new experiments to address some of the issues aforementioned and determine the impact they have on retrieval performance.
Research e orts on GIR are addressing issues such as access to multilingual documents, techniques
for information mining (i.e., extraction, exploration and visualization of geo-referenced
information), investigation of spatial representations and ranking methods for di erent representations,
application of machine learning techniques for place name recognition, development of datasets
containing annotated geographic entities, among others. [
The focus of one of the tasks was on experimenting with and evaluating the performance of GIR systems when topics include geographic references. Collections of documents and topics in di erent languages were available to carry out monolingual and bilingual experiments. We ran monolingual experiments in English, German, and Portuguese; for bilingual retrieval, we worked with topics in German and English and collections in English, German and Portuguese.
In this paper we describe experiments in the cross-language monolingual and bilingual task. We used the Terrier Information Retrieval (IR) platform version 2.1 to run our experiments. This platform has performed successfully in monolingual information retrieval tasks in CLEF and TREC. The paper is organized as follows. In Section 2 we present our work in the monolingual task including an overview of Terrier. Section 3 describes our setting and experiments in the bilingual task. Finally, we present conclusions and current work in Section 4. 2
Cross-lingual Geographical IR Task In this section we present Terrier (TERabyte RetRIEveR) an information retrieval (IR) platform used in all the experiments. Then we describe experiments and results for monolingual GIR in English, German, and Portuguese. The nal subsection includes the experiments and results for bilingual GIR with topics in German and English.
Terrier is a high performance and scalable search engine platform for the rapid development of
large-scale retrieval applications. It o ers a variety of IR models based on the Divergence from
Randomness (DFR) framework ([
The components of the DFR models are the following: 1) a randomness model; 2) an information gain model; and 3) a term frequency normalization model. The latter component adjusts the frequency of a term in a document based on the length of a document and the average document length in the entire collection. For example, the Normalization 2 term frequency normalization model assumes a decreasing density function of the normalized term frequency concerning the document length.
The normalized term frequency tfn is calculated as follows:
tf n = tf:log2(1 + c
avg len
len
)
tf is the term frequency, avg len is the average document length in the collection, and len is the
document length, and c is a hyper-parameter. We used c = 1.5 for short queries, which is the
default value, c = 3.0 for short queries. Short queries in our context are those which use only the
topic title eld and the topic description eld. We used these values based on the results generated
by the experiments on tuning for BM25 and DFR models done by He and Ounis [
Each query term in a document is assigned a weight depending how important the term is to the document. Term weights are then used to match documents to a query. Documents are ranked according to their estimated relevance to the query. The formula to estimate the probability of producing the query for a given document is the sum of the probability of producing the terms in the query plus the probability of not producing other terms.
2.1
ald (British) 1995 for English, publico94, publico95, folha94 and folha95 for Portuguese, and der spiegel, frankfurter and fr rundschau for German. There were 25 topics for each of the languages tested. Documents and topics were processed using the English stopwords list and the
2.2
Run Id
Rel. Ret. Run Id
Rel. Ret.
Unlike the monolingual runs and the Spanish !English run, relevance feedback did not improve performance retrieval. No querying was done with the language model option. 4
Conclusions
In this paper we presented work on monolingual and bilingual geographical information retrieval.
We used Terrier to run our experiments using the InL2 parameter-based model. Comparing results
with those obtained in the past three years (see [