=Paper=
{{Paper
|id=Vol-1171/CLEF2005wn-CLSR-LamAdesinaEt2005
|storemode=property
|title=Dublin City University at CLEF 2005: Cross-Language Spoken Document Retrieval (CL-SR) Experiments
|pdfUrl=https://ceur-ws.org/Vol-1171/CLEF2005wn-CLSR-LamAdesinaEt2005.pdf
|volume=Vol-1171
|dblpUrl=https://dblp.org/rec/conf/clef/Lam-AdesinaJ05b
}}
==Dublin City University at CLEF 2005: Cross-Language Spoken Document Retrieval (CL-SR) Experiments==
https://ceur-ws.org/Vol-1171/CLEF2005wn-CLSR-LamAdesinaEt2005.pdf
Dublin City University at CLEF 2005: Cross-Language Spoken Document
Retrieval (CL-SR) Experiments
Adenike M. Lam-Adesina, Gareth J. F. Jones
School of Computing, Dublin City University, Dublin 9, Ireland
{adenike,gjones}@computing.dcu.ie
Abstract
The Dublin City University participation in the CLEF CL-SR 2005 task concentrated on exploring the
application of our existing information retrieval methods based on the Okapi model to the conversational speech
data set. This required an approach to determining approximate sentence boundaries within the free-flowing
automatic transcription provided. We also performed exploratory experiments on the use of the metadata
provided with the document transcriptions. Topics were translated into English using Systran V3.0 machine
translation.
Categories and Subject Descriptors
H.3 Information Storage and Retrieval]: H.3.1 Content Analysis and Indexing; H.3.3 Information Search and
Retrieval - Relevance Feedback; H.3.7 Digital Libraries
General Terms
Measurement, Performance, Experimentation
Keywords
Cross-language spoken document retrieval, Transcription segmentation, Pseudo relevance feedback, Metadata
combination
1 Introduction
The Dublin City University participation in the CLEF CL-SR 2005 task concentrated on exploring the
application of our existing information retrieval methods based on the Okapi model for this data set, and
exploratory experiments on the use of the provided document metadata. Our official submissions included both
the English monolingual and French bilingual runs. This paper reports additional results for German and Spanish
bilingual runs. Topics were translated into English using Systran V3.0 machine translation system. The resulting
English topics were used for retrieving from the English document collection.
Our standard Okapi retrieval system incorporates a summary-based pseudo relevance feedback (PRF)
stage. This PRF system operates by selecting topic expansion terms from document summaries, full details are
described in [1]. However, since the automated transcriptions of the conversational speech documents do not
contain punctuation, we needed to develop a method of selecting significant document segments. Details of this
method are described in Section 2.1.
The document transcriptions are provided with a rich set of metadata. It is not immediately clear how
best to exploit this most effectively in retrieval. This paper reports our initial exploratory experiments in making
use of this additional information.
The remainder of this paper is structured as follows: Section 2 overviews our retrieval system and
describes our sentence boundary creation technique, Section 3 presents the results of our experimental
investigations, and Section 4 concludes the paper with a discussion of our results.
2 System Setup
The basis of our experimental system is the City University research distribution version of the Okapi system [2].
The documents and search topics were processed to remove stopwords from a list of about 260 words, suffix
stripped using the Okapi implementation of Porter stemming [3] and terms were indexed using a small standard
set of synonyms. None of these procedures were adapted for the new CL-SR document set.
2.1 Term Weighting
Document terms were weighted using the Okapi BM25 weighting scheme developed in [2] calculated as follows,
cfw (i ) × tf (i , j ) × ( K 1 + 1)
cw (i , j ) = (1)
K 1 * ((1 − b ) + (b × ndl ( j ))) + tf (i , j )
where cw(i,j) represents the weight of term i in document j, cfw(i) is the standard collection frequency weight,
tf(i,j) is the document term frequency, and ndl(j) is the normalized document length. ndl(j) is calculated as ndl(j)
= dl(j)/avdl where dl(j) is the length of j and avdl is the average document length for all documents. k1 and b are
empirically selected tuning constants for a particular collection. k1 is designed to modify the degree of effect of
tf(i,j), while constant b modifies the effect of document length. High values of b imply that documents are long
because they are verbose, while low values imply that they are long because they are multi-topic. Our values are
tuned based on the training topics.
2.2 Pseudo-Relevance Feedback
We apply PRF for query expansion using a summary-based method described in [1] which has been shown to be
effective in our previous submissions to CLEF, including [4] and elsewhere. The main challenge for query
expansion is the selection of appropriate terms from the assumed relevant documents. Our query expansion
method selects terms from summaries of the top ranked relevant document. All non-stopwords in the summaries
are ranked using a slightly modified version of the Robertson selection value (rsv) [] shown in equation (2).
rsv(i) = r(i) × rw(i) (2)
where r(i) = number of relevant documents containing term i, and rw(i) is the standard Robertson/Sparck Jones
relevance weight [2],
( r (i ) + 0.5)( N − n(i ) − R + r (i ) + 0.5)
rw(i ) = log
( n(i ) − r (i ) + 0.5)( R − r (i ) + 0.5)
where n(i) = the total number of documents containing term i, r(i) = the total number of relevant documents term
i occurs in, R = the total number of relevant documents for this query, and N = the total number of documents
The top ranked terms are then added to the topic. In our modified version of the rsv(i), potential
expansion terms are selected from the summaries of the top ranked documents, but ranked using statistics from
top a large number of assumed relevant ranked documents from the initial run.
2.2.1 Sentence Selection
Our standard process for summary generation is to select representative sentences from the document. Since the
document transcriptions do not contain punctuation marking we needed an alternative approach to identifying
significant units in the transcription. We approached this using a method derived from Luhn’s word cluster
hypothesis. Luhn’s hypothesis states that significant words separated by not more than 5 non-significant words
are strongly related. Strongly related word clusters were identified in the running document transcription by
searching for word groups separated by not more than 5 insignificant words, as shown in Figure 1. Note that
words appearing between clusters are not included in clusters, but can be ignored for the purposes of query
expansion since they are by definition stop words.
… this chapter gives a brief description of the [data sets used in evaluating the automatic relevance
feedback procedure investigated in this thesis] and also discusses the extension of …
Fig 1. Example of Sentence creation.
The clusters were then awarded a significance score based on two measures.
Luhn’s Keyword Cluster Method Luhn‘s method assigns a sentence score for highest scoring cluster within a
sentence. We adopted this method to assign a cluster score as follows:
SW 2
SS1 =
TW
where SS1 = the sentence score
SW = the number of bracketed significant words (in this case 6)
TW = the total number of bracketed words (in this case 14)
Query-Bias Method This method assigns a score to each sentence based on the number of query terms in the
sentence as follows:
TQ 2
SS 2 =
NQ
where SS2 = the sentence score
TQ = the number of query terms present in the sentence
NQ = the number of terms in a query
The overall score for each sentence (cluster) was then form by summing these two measures for each sentence.
3 Experimental Investigation
This section describes the establishment of the parameters of our experimental system and gives results from our
investigations.
3.1 Selection of System Parameters
In order to set the appropriate parameters for our feedback runs, we carried out development runs using the
CLSR 2005 training topics. The Okapi parameters were set as follows k1=1.4 b=0.8. For all our PRF runs, 5
documents were assumed relevant for term selection and document summaries comprised the best scoring 4
clusters. The rsv values to rank the potential expansion terms were estimated based on the top 20 or 40 ranked
assumed relevant documents. The top 20 ranked expansion terms taken from the clusters were added to the
original query in each case. Based on results from our previous experiments in CLEF, the original topic terms are
up-weighted by a factor of 3.5 relative to terms introduced by PRF. For our runs we used either the Title section
(dcu*tit) or the Title and Description (dcu*desc) section of each topic. Our official runs are marked *. Initial
baseline monolingual results using English without query expansion are also given for comparison.
For our experiments the document fields were combined as follows,
dcua2 – combination of ASRTEXT2004A and AUTOKEYWORDA1
dcua1a2 – combination of ASRTEXT2004A, AUTOKEYWORDA1 and AUTOKEYWORDA2
dcusum – combination of ASRTEXT2004A, AUTOKEYWORDA1 and AUTOKEYWORDA2 and the
SUMMARY
dcuall – combination of NAME, MANUALKEYWORD, SUMMARY and ASRTEXT2004A section of each
documents.
3.2 Experimental Results
Tables 1-4 show results of our experiments using these different data combinations for the 25 test topics released
for the CLEF 2005 CL-SR task. Results shown are Mean Average Precision (MAP), total relevant retrieved (Rr),
and precision at cutoffs of 10 and 30 documents. Topic languages used are English, French, German and
Spanish. Topics were translated into English using the Systran V3.0 machine translation system. The upper set of
results in each table shows combined Title and Description topic queries and the lower set Title only topic
queries.
Run-id Topic Lang. MAP Rr P10 P30
dcua2desc40f Baseline 0.0496 536 0.1480 0.1027
English 0.0654* 738 0.1760 0.1400
French 0.0756 744 0.2080 0.1387
German 0.0407 611 0.1160 0.0987
Spanish 0.0549 727 0.1520 0.1093
dcua2tit40f Baseline 0.0703 384 0.2280 0.1427
English 0.0795 622 0.2520 0.1507
French 0.0805 708 0.2520 0.1547
German 0.0555 647 0.1840 0.1200
Spanish 0.0681 602 0.1920 0.1293
Table 1: Results using a combination of ASRTEXT2004A and AUTOKEYWORDA1, the Title or Title and
Description topic fields. Expansion terms ranked for selection using statistics of 40 top ranked documents.
Results in Table 1 show results for combination of ASRTEXT2004A with AUTOKEYWORDA1. It can be seen
that the PSF method improves results for the English topics. Also that the results using Title only topics is better
than using the combined Title and Description topics with respect to MAP. This result is perhaps a little
surprising since the latter are generally found to be perform better and we are investigating the reasons for the
results observed here. However, the number of relevant documents retrieved is generally higher when using the
combined topics. Cross language results using French topics are shown to perform better than monolingual
English for both MAP and relevant retrieved. This is again unusual, but not unprecedented. Results for translated
German and Spanish show a reduction on the monolingual results.
Run-id Topic Lang. MAP Rr P10 P30
dcua1a2desc40f Baseline 0.0464 500 0.1880 0.1053
English 0.0670 784 0.1840 0.1480
French 0.0943 773 0.2160 0.1707
German 0.0455 611 0.0960 0.0920
Spanish 0.0640 765 0.1640 0.1280
dcua1a2tit40f Baseline 0.0796 472 0.2280 0.1600
English 0.1101* 727 0.2520 0.1960
French 0.1064* 768 0.2600 0.1907
German 0.0740 691 0.1720 0.1493
Spanish 0.0912 679 0.2200 0.1560
Table 2: Results using a combination of ASRTEXT2004A, AUTOKEYWORDA1 and AUTOKEYWORDA2,
the Title or Title and Description topic fields. Expansion terms ranked for selection using statistics of 40 top
ranked documents.
Table 2 shows results for the same experiments to those in Table 1 with the addition of AUTOKEYWORDA2 to
the documents. Results here are generally show similar trends to those in Table 1 with small absolute increases
in performance in most cases. In this case the performance advantage of French topics over English topics with
PRF has largely disappeared.
Run-id Topic Lang. MAP Rr P10 P30
dcusumdesc40f Baseline 0.1047 598 0.2240 0.1707
English 0.1472 889 0.2720 0.2173
French 0.1544 856 0.2600 0.2160
German 0.1083 696 0.1640 0.1373
Spanish 0.1074 860 0.1680 0.1520
Dcusumtit40f Baseline 0.1407 618 0.2840 0.2160
English 0.1672 770 0.2920 0.2427
French 0.1654* 837 0.3080 0.2507
German 0.1100 738 0.2200 0.1600
Spanish 0.1542 736 0.2840 0.1296
Table 3: Results using a combination of ASRTEXT2004A, AUTOKEYWORDA1 and AUTOKEYWORDA2
and the SUMMARY section of each document, the Title or Title and Description topic fields. Expansion terms
ranked for selection using statistics of 40 top ranked documents.
Table 3 shows results for a further set of experiments with the SUMMARY field added to the document
descriptions. Results here show large increases compared to those in Table 2, indicating that the contents of the
SUMMARY field are useful descriptions of the documents.
Run-id Topic Lang. MAP Rr P10 P30
Dcualldesc40f Baseline 0.2213 1031 0.3680 0.2707
English 0.3073 1257 0.4880 0.3773
French 0.2762 1122 0.4960 0.3600
German 0.2045 1001 0.3600 0.2760
Spanish 0.2320 1160 0.3600 0.2680
Dcualltit40f Baseline 0.2421 736 0.4120 0.3107
English 0.2833 1009 0.4320 0.3373
French 0.2569 1136 0.4240 0.3027
German 0.2288 962 0.3280 0.2720
Spanish 0.2468 908 0.3800 0.2973
Table 4: Results using a combination of NAME, MANUALKEYWORD, SUMMARY and ASRTEXT2004A
section of each document, the Title or Title and Description topic fields. Expansion terms ranked for selection
using statistics of 40 top ranked documents.
Table 4 shows a final set of experiments combining the NAME, MANUALKEYWORD and SUMMARY fields
with ASRTEXT2004A. These results show large improvements over the results shown in previous tables.
Performance for Title only and Title and Description combined topics is now similar with neither clearly
showing an advantage. Monolingual English performance is clearly better than results for translated French
topics, while our PRF method is still shown to be effective.
4 Conclusions and Further Work
Our initial experiments with the CLEF 2005 CL-SR task illustrate a number of points: PRF can be successfully
applied to this data set, that the different fields of the document set make varying contributions to information
retrieval effectiveness. In general in can be seen that manual assigned fields are more usefully than the
automatically generated ones.
These experiments only represent a small subset of those that are possible with this dataset. In order to
better understand the usefulness of document fields and retrieval methods more detailed analysis of these
existing results and further experiments are planned.
References
[1] A.M. Lam-Adesina and G.J.F. Jones. Applying Summarization Techniques for Term Selection in
Relevance Feedback. In Proceedings of the 24th Annual International ACM SIGIR Conference on
Research and Development in Information Retrieval, pages 1-9, New Orleans, 2001. ACM.
[2] S.E Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu and M. Gatford, Okapi at TREC-3. In D.K.
Harman, editor, Proceedings of the Third Text REtrieval Conference (TREC-3), pages 109-126. NIST,
1995.
[3] M.F. Porter. An algorithm for suffix stripping. Program, 14:10-137, 1980. H.P. Luhn. The Automatic
Creation of Literature Abstracts. IBM Journal of Research and Development, 2(2):159-165, 1958.
[4] G.J.F.Jones, M.Burke, J.Judge, A.Khasin, A.Lam-Adesina and J.Wagner ,Dublin City University at CLEF
2004: Experiments in Monolingual, Bilingual and Multilingual Retrieval, Proceedings of the CLEF 2004:
Workshop on Cross-Language Information Retrieval and Evaluation, Bath, U.K., 2004.
[5] A. Tombros and M. Sanderson. The Advantages of Query-Biased Summaries in Information Retrieval. In
proceedings of the 21st Annual International ACM SIGIR Conference Research and Development in
Information Retrieval, pages 2-10, Melbourne, 1998. ACM.