The Options Panel is a GUI panel allowing users to select options affecting the details of feature extraction. This panel is divided into two sections. The left section allows users to select the web service(s) to submit queries to and to set query preferences that affect the types and numbers of hits returned by queries. The right section allows users to set the details of how final feature values will be calculated from raw hit counts and which reports will be displayed in the Results Panel after feature extraction is complete. A screenshot of the Options Panel is shown in Figure 1.

Figure 1: A screenshot of the Options Panel showing default settings.

This section of the manual is divided into seven sections, one for each of the seven sections of the Options Panel interface:

1) WEB SERVICES TO SEARCH

The options in this section allow the user to control which web services to submit queries to. Passwords to use the services may also be entered here, if appropriate.

Although multiple web services may be selected, this will increase the duration of feature extraction compared to only using one web service. The hit counts for each selected web service are combined during final feature calculation. The limitations of each web service are described in the Hints and Suggestions section of this manual.

It may be useful for those with software development backgrounds to implement additional web services, as described in the Extending the Software section of this manual.

The specific options in this section are as follows:

• Yahoo!: Whether queries are to be submitted to the Yahoo! REST-like Web Search web services. The Yahoo! Application Key is hard coded into jWebMiner because Yahoo!'s quota is 5000 queries per day per source IP address, so multiple users will not reduce the quota available to each user.
• Google: Whether queries are to be submitted to the Google SOAP web services. A Google distributed License Key must be entered in the License key field for Google to be accessed. This key is not hard coded into jWebMiner because Google imposes a limit of 1000 queries per day with a particular key, regardless of source IP address, so each user needs their own key to avoid reducing the quota available to other users.

2) GENERAL SEARCH SETTINGS

The options in this section allow the user to control miscellaneous settings influencing how searches are performed and what types of results are returned.

It is in general best for users to leave these preferences as they are unless they are expert users. This is because these controls can have significant effect on returned hit counts and because not all of these options are supported by all web services.

The specific options in this section are as follows:

• Treat strings literally: Whether all search queries should be literal searches (e.g. for the query "heavy metal" sites must have the two words adjacent if they are to be considered a hit if the search is literal). This is also sometimes known as an exact search or a phrase search.
• Perform search as OR instead of AND: Whether search queries need only contain one of the specified query words in order to result in a hit. If this is set to true, then only one of the query words must be present. If this is set to false, then all of them must be present (although not necessarily in the specified order, unless the Treat strings literally checkbox is selected). It is recommended that the user leave this box unchecked.
• Include non-matching similar hits: Whether results returned by search queries may include hits that do not contain one or more of the specified query words but do contain terms very similar to them (e.g. alternative spellings).
• Suppress similar hits: Whether to suppress similar hits when reporting results. Similar in this context means either sites with identical titles and/or descriptions, or multiple hits from the same host.
• Suppress adult content: Whether to suppress hits that are classified as containing adult content by the search service in question.

3) LANGUAGE, REGIONAL AND FILE TYPE FILTERS

The options in this section allow the user to control filters (of a different type than those in the Required Filter Words Panel and Excluded Filter Terms Panel) that limit the types of hits that can be returned. The geographical region that searches are performed in can also be specified. These options must be chosen from the available options in the provided combo boxes. A choice of "No Limitations" means that the corresponding filter is not applied.

The specific options in this section are as follows:

• Limit to language: Sets the name of a language that pages must be in in order to be counted as hits.
• Limit to country: Sets the country that pages must be found in in order to be counted as hits.
• Search from region: Sets the name of a country where searches will be performed (i.e. where the search service is located). Results are not necessarily limited to this country, however.
• Limit to file type: Sets the file type that documents must be in in order to be counted as hits.

4) FEATURE SCORE CALCULATION SETTINGS

The options in this section control whether various types of normalization are applied during final feature calculation. None, some or all of these normalizations can be applied. The specific options in this section are as follows:

• Normalize hits across web services: Whether hit counts from different web services should be normalized before final feature scores are calculated. This ensures that services that produce fewer hits overall are not underweighted in the final scores. The overall hit counts for each web service are scaled to the overall number of hits of the service with the lowest number of hits. This checkbox has no effect if only one web service is selected in the WEB SERVICES TO SEARCH section of the Options Panel.
• Normalize hits across web sites: Whether hits from different sources (individual web sites and/or the network as a whole, as set in the Site Weightings Panel) should be normalized before final feature scores are calculated. This ensures that sources that produce fewer hits overall are not underweighted in the final scores. This is separate and in addition to the manual weights that can be imposed on the Site Weightings Panel, which are not affected by whether or not normalization is applied. The overall hit counts for each source are normalized to the overall number of hits of the source with the lowest number of hits overall. This checkbox has no effect if only one source is specified with the Site Weightings Panel.
• Normalize feature settings: Whether final feature values should be normalized after all other feature calculation steps have been completed. If the By row rather than overall checkbox is selected then this normalization is performed on a row by row basis (so that the sum of feature values in each row of the table of feature scores is 1.0), otherwise it is an overall normalization for the table of scores (so that the sum of all feature values is 1.0).

5) CO-OCCURRENCE SCORING FUNCTION

The section controls which formula is used to calculate final feature values from hit counts when the Co-Occurrence Extraction option is selected in the Search Words Panel. These formulas are applied to hit counts after hit counts have been combined across sources (as set in the Site Weightings Panel) and across web services (as set in the WEB SERVICES TO SEARCH section of the Options Panel). These formulas are also applied after source weightings (as set in the Site Weightings Panel) and normalizations (as set in the FEATURE SCORE CALCULATION SETTINGS section of the Options Panel) have been used to process hit counts. The exception to this is the Normalize feature settings normalization, which is applied after the chosen formula has been applied.

Which formula is best to use depends on the kind of search that is being performed, and it can be useful to experiment with different formulas. It may also be useful for those with software development backgrounds to implement additional formulas, as described in the Extending the Software section of this manual.

The specific options in this section are given below as well as references to publications that detail their previous use.

• S(a1,a2) = C(a1,a2) / (1 + (SUMcfromAcnota1(C(a1,c)) x SUMdfromAdnota2(C(d,a2))))
  • Geleijnse, G., and J. Korst. 2006. Web-based artist categorization. Proceedings of the International Conference on Music Information Retrieval. 266–71.
• S(a1,a2, c) = (C(a1,a2) / C(a2)) * (1 - (|C(a1) - C(a2)| / C(c)))
  • Ellis, D. P. W., B. Whitman, A. Berenzweig, and S. Lawrence. 2002. The quest for ground truth in musical artist similarity. Proceedings of the International Conference on Music Information Retrieval. 170–7.
  • In this formula c refers to the the element from A with the largest C(A).

The notation used in the above formulas is defined as follows:

• S() refers to the scoring function that outputs feature scores (before Normalize feature settings normalization, if selected).
• C() refers to the processed hit counts for the input query combined across web services and weighted network sites and after web service and web site normalization, if selected. If only one web service is used and only one source (including the possibility of the whole network) is used then this is just the raw hit count for the input query.
• A is the set of lines in the PRIMARY SEARCH STRINGS field of the Search Words Panel.
• a1 and a2 each refer to individual elements from the set A.

6) CROSS TABULATION SCORING FUNCTION

The section controls which formula is used to calculate final feature values from hit counts when the Cross Tabulation Extraction option is selected in the Search Words Panel. These formulas are applied to hit counts after hit counts have been combined across sources (as set in the Site Weightings Panel) and across web services (as set in the WEB SERVICES TO SEARCH section of the Options Panel). These formulas are also applied after source weightings (as set in the Site Weightings Panel) and normalizations (as set in the FEATURE SCORE CALCULATION SETTINGS section of the Options Panel) have been used to process hit counts. The exception to this is the the Normalize feature settings normalization, which is applied after the chosen formula has been applied.

Which formula is best to use depends on the kind of search that is being performed, and it can be useful to experiment with different formulas. It may also be useful for those with software development backgrounds to implement additional formulas, as described in the Extending the Software section of this manual.

The specific options in this section are given below as well as references to publications that detail their previous use.

• S(a,b) = C(a,b) / C(a)
  • Schedl, M., T. Pohle, P. Knees, and G. Widmer. 2006. Assigning and visualizing music genres by web-based co-occurrence analysis. Proceedings of the International Conference on Music Information Retrieval. 260–5.
• S(a,b) = C(a,b) / C(b)
  • Schedl, M., T. Pohle, P. Knees, and G. Widmer. 2006. Assigning and visualizing music genres by web-based co-occurrence analysis. Proceedings of the International Conference on Music Information Retrieval. 260–5.
• S(a,b) = C(a,b) / (1 + SUMcfromA(C(c,b)))
  • Geleijnse, G., and J. Korst. 2006. Web-based artist categorization. Proceedings of the International Conference on Music Information Retrieval. 266–71.
• S(a,b) = C(a,b) / (1 + SUMcfromB(C(a,c)))

The notation used in the above formulas is defined as follows:

• S() refers to the scoring function that outputs feature scores (before Normalize feature settings normalization, if selected).
• C() refers to the processed hit counts for the input query combined across web services and weighted network sites and after web service and web site normalization, if selected. If only one web service is used and only one source (including the possibility of the whole network) is used then this is just the raw hit count for the input query.
• A is the set of lines in the PRIMARY SEARCH STRINGS field of the Search Words Panel.
• B is the set of lines in the SECONDARY SEARCH STRINGS field of the Search Words Panel.
• a refers to an individual element from the set A.
• b refers to an individual element from the set B.

7) INFORMATION TO REPORT

The options in this section control which types of reports are generated and displayed in the Results Panel after feature extraction is complete. These reports are each presented in tables labeled by search strings (except for the Search settings used report).

All, some or none of these reports may be selected, but regardless of the reports selected here feature values are also stored after extraction so that they can be saved in the Results Panel as ACE XML, Weka ARFF or newline delimited text files. Reports are displayed in the Results Panel in the same order that they appear here in the Options Panel. These reports can be useful in debugging and/or understanding why feature scores are as they are, as they can be used to view processing at various intermediate stages.

Note that in the case where the Co-Occurrence Extraction option is selected in the Search Words Panel that entries on the diagonal of each report table are left empty in each of the reports and are set to 0 if saved, as they are not significant.

The specific options in this section are as follows:

• Feature scores: Whether or not to display the final feature scores that may be saved as features after feature extraction is complete. This report is displayed in a table with the highest value in each row italicized and bolded.
• Combined processed hit counts: Whether or not to display the combined (added) processed hit counts for all web services and sites after pre-processing is complete. This does take into account site weighting and/or source and web service normalization, if they have been selected by the user. It does not include the application of scoring functions or feature score normalization, however.
• Combined raw hit counts: Whether or not to display the combined hit counts added together across all web services and sites after queries are all completed. This does not take into account site weighting, any normalizations or application of scoring functions.
• Individual raw hit counts: Whether or not to individually display the raw hit counts for each web service and each network source after queries are all completed but before hit counts have been combined or otherwise processed.
• Search queries used: Whether or not to display the queries that are actually sent to each of the web services. Note that these queries will not include information that was stored in the actual web service objects themselves, so the queries may appear incomplete in the case of some web services. This report is combined into the same table with the Individual raw hit counts report if it is selected to be generated, and is generated separately if it is not.
• Search settings used: Whether or not to display the details of all of the settings used to perform feature extractions. This is useful for maintaining a record of the settings used each time that features are extracted.

-top of page-