Chapter 8. Searching documents

Table of Contents
8.1. Using search front-ends
8.2. mod_dpsearch module for Apache httpd
8.3. How to write search result templates
8.4. Designing search.html
8.5. Relevance
8.6. Search queries tracking
8.7. Search results cache
8.8. Fuzzy search

8.1. Using search front-ends

8.1.1. Performing search

Open your preferred front-end in Web browser:

http://your.web.server/path/to/search.cgi

To find something just type words you want to find and press SUBMIT button. For example: mysql odbc. DataparkSearch will find all documents that contain word mysql and/or word odbc. Best documents having bigger weights will be displayed first.

To find a phrase, simple enclose it in quotas. For example: "uncontrollable sphere".

8.1.2. Search parameters

DataparkSearch front-ends support the following parameters given in CGI query string. You may use them in HTML form on search page.

Table 8-1. Available search parameters

qtext parameter with search query
vqtext parameter with search query in the Verity Query Language (prefix variant), see Section 8.1.8>. To use this parameter, you need to leave empty the q parameter.
scharacters sequence, specify results sorting order. Small caps specify ascendant sorting, upper caps - descendant. Following characters can be used: R or r - for sorting by relevance, P or p - for sorting by PopularityRank, I or i - for sorting by Importance (multiplication of relevance and PopularityRank), A or a - for sorting by sum of relevance and PopularityRank, D or d - for sorting by last modified date. Default value: RP.
pspage size, number of search results displayed on one page, 20 by default. Maximum page size is 100. This value does not allow passing very big page sizes to avoid server overload and might be changed with MAX_PS definition in search.c.
nppage number, starting by 0, 0 by default (first page)
ppage number starting by 1. Suitable for use with OpenSearch
msearch mode. Currently "all","any", "near" and "bool" values are supported.
wmword match. You may use this parameter to choose word match type. There are "wrd", "beg", "end" and "sub" values that respectively mean whole word, word beginning, word ending and word substring match.
tTag limit. Limits search through only documents with given tag. This parameter has the same effect with -t indexer option
cCategory limit. Take a look into Section 6.2> for details.
ulURL limit, URL substring to limit search through subsection of database. It supports SQL % and _ LIKE wildcards. This parameter has the same effect with -u indexer option. If relative URL is specified search.cgi inserts % signs before and after "ul" value when compiled with SQL support. It allows to write URL substring in HTML from to limit search, for example <OPTION VALUE="/manual/"> instead of VALUE="%/manual/%". When full URL with schema is specified search.cgi adds % sign only after this value. For example for <OPTION VALUE="http://localhost/"> search.cgi will pass http://localhost/% in SQL LIKE comparison.
Not supported for cache storage mode.
wfWeight factors. It allows changing different document sections weights at a search time. Should be passed in the form of hex number. Check the explanation below.
gLanguage limit. Language abbreviation to limit search results by url.lang field.
tmpltTemplate filename (without path). To specify template file other standard search.htm.
typeContent-Type limit. Content-type to limit search results by url.content_type field. For cache mode storage this should be exact match. For SQL-modes it may be sql-like pattern.
spWords forms limit. =1, if you need search all forms (include spelling suggestions, if aspell support is enabled) for entered words. =0, if you need search only entered words. Default value is 1. You may switch it to 0 for faster search.
sySynonyms limit. =1, if you need add synonyms for entered words. =0, do not use synonyms. Default value is 1. You may switch it to 0 for faster search.
emptyUse limits to show results if no query words is entered (only for cache mode). =yes, to show results from limits, if no query words is entered (default). =no, do not show results from limits, if no query words is entered.
dtLimit by time. Three types is supported.

If dt is set to back, that means you want to limit result to recent pages, and you should specify this recentness in variable dp.

If dt is set to er, that means the search will be limited to pages newer or older than date given. Variable dx is newer/older flag (1 means newer or after, -1 means older or before). Date is specified in variables dm, dy, dd.

If dt is set to range, that means search within given range of dates. Variables db and de are used here and stands for beginning and end date.

All times in cache mode measured in a hour precision.
dpLimit by recentness, if dt value is back. It should be specified in xxxA[yyyB[zzzC]] format. Spaces are allowed between xxx and A and yyy and so on). xxx, yyy, zzz are numbers (can be negative!), A, B, C can be one of the following (the letters are the same as in strptime/strftime functions): s - second, M - minute, h - hour, d - day, m - month, y - year. Examples:
  4h30M 	  - 2 hours and 30 minutes
  1Y6m-15d        - 1 year and six month minus 15 days
  1h-60M+1s       - 1 hour minus 60 minutes plus 1 second
dxis newer/older flag (1 means newer or after, -1 means older or before), if dt value is er.
dmMonth, if dt value is er. 0 - January, 1 - February, ... 11 - December.
dyYear, if dt value is er. Four digits. For example, 1999 or 2001.
ddDay, if dt value is er. 1...31.
dbBeginning date, if dt value is range. Each date is string in the form dd/mm/yyyy, there dd is day, mm is month and yyyy is four-digits year.
deEnd date, if dt value is range. Each date is string in the form dd/mm/yyyy, there dd is day, mm is month and yyyy is four-digits year.

8.1.3. Changing different document parts weights at search time

It is possible to pass wf HTML form variable to search.cgi. wf variable represents weight factors for specific document parts. Currently body, title, keywords, description, url parts, crosswords as well as user defined META and HTTP headers are supported. Take a look into "Section" part of indexer.conf-dist.

To be able use this feature it is recommended to set different sections IDs for different document parts in "Section" indexer.conf command. Currently up to 256 different sections are supported.

Imagine that we have these default sections in indexer.conf:

  Section body        1  256
  Section title       2  128
  Section keywords    3  128
  Section description 4  128

wf value is a string of hex digits ABCD. Each digit is a factor for corresponding section weight. The most right digit corresponds to section 1. For the given above sections configuration:

      D is a factor for section 1 (body)
      C is a factor for section 2 (title)
      B is a factor for section 3 (keywords)
      A is a factor for section 4 (description)

Examples:

   wf=0001 will search through body only.

   wf=1110 will search through title,keywords,description but not 
through the body.

   wf=F421 will search through:
          Description with factor 15  (F hex)
          Keywords with factor 4
          Title with factor  2
          Body with factor 1

By default, if wf variable is omitted in the query, all section factors are 1, it means all sections have the same weight. If the number of sections in wf is less than the number of sections defined, then the rest sections are initialized by the value of highest section weight defined in wf. E.g.:

   wf=01 will also search through body only.

If DataparkSearch has been built with fast relevance calculation (with --enable-rel=fast option for configure), in this case, only zero and non-zero values for wf variable take an effect (this allows only include/exclude specified sections in search results). To use full support for dynamic section weights, you need specify --enable-rel=full option for configure when configuring DataparkSearch.

8.1.4. Using front-end with an shtml page

When using a dynamic shtml page containing SSI that calls search.cgi, i.e. search.cgi is not called directly as a CGI program, it is necessary to override Apache's SCRIPT_NAME environment attribute so that all the links on search pages lead to the dynamic page and not to search.cgi.

For example, when a shtml page contains a line <--#include virtual="search.cgi">, SCRIPT_NAME variable will still point to search.cgi, but not to the shtml page.

To override SCRIPT_NAME variable we implemented a DPSEARCH_SELF variable that you may add to Apache's httpd.conf file. Thus search.cgi will check DPSEARCH_SELF variable first and then SCRIPT_NAME. Here is an example of using DPSEARCH_SELF environment variable with SetEnv/PassEnv Apache's httpd.conf command:

SetEnv DPSEARCH_SELF /path/to/search.cgi
PassEnv DPSEARCH_SELF

8.1.5. Using several templates

It is often required to use several templates with the same search.cgi. There are actually several ways to do it. They are given here in the order how search.cgi detects template name.

  1. search.cgi checks environment variable DPSEARCH_TEMPLATE. So you can put a path to desired search template into this variable.

  2. search.cgi checks path info part of URL available in the PATH_INFO environment variable. E.g. http://localhost/cgi-bin/search.cgi/search1.html uses search1.htm as its template, and http://localhost/cgi-bin/search.cgi/search2.html uses search2.htm, and so on.

  3. search.cgi also supports Apache internal redirect. It checks REDIRECT_STATUS and REDIRECT_URL environment variables. To activate this way of template usage you may add these lines in Apache srm.conf:

    AddType text/html .zhtml
    AddHandler zhtml .zhtml
    Action zhtml /cgi-bin/search.cgi

    Put search.cgi into your /cgi-bin/ directory. Then put HTML template into your site directory structure under any name with .zthml extension, for example template.zhtml. Now you may open search page: http://www.site.com/path/to/template.zhtml You may use any unused extension instead of .zthml of course.

  4. If the above two ways fail, search.cgi opens a template which has the same name with the script being executed using SCRIPT_NAME environment variable. search.cgi will open a template ETC/search.htm, search1.cgi will open ETC/search1.htm and so on, where ETC is DataparkSearch /etc directory (usually /usr/local/dpsearch/etc). So, you can use the same search.cgi with different templates without having to recompile it. Just create one or several hard or symbolic links for search.cgi or copy it and put corresponding search templates into /etc directory of DataparkSearch installation.

    Take a look also into Making multi-language search pages section

8.1.6. Search operators

The operator allin<section>:, where <section> is the name of a section, defined in sections.conf file (or in any dpsearch's configuration file by Section command) with non-zero section number (see Section 3.10.43>), that operator allows to limit the search domain for a query word by the section specified.

This operator differ from limiting search domain using &wf= CGI-variable in a way, that such limit is imposing only on query words specified after this operator.

For example, if you have the following commands in sections.conf file

Section body 1 256
Section title 2 128
Section url 3 0 strict
then you can use the following operators in search query: allinbody:, allintitle: and allinurl:.

For the query computer allintitle: science it will be found the documents that contain the word "science" in the title and the word "computer" in any document section.

8.1.7. Advanced boolean search

If you want more advanced results you may use query language. You should select "bool" search mode in the search from.

DataparkSearch understands the following boolean operators:

AND or & - logical AND. For example, "mysql & odbc" or "mysql AND odbc" - DataparkSearch will find any URLs that contain both "mysql" and "odbc".

NEAR - NEAR operator, identical to AND operator, but come true if both words are within 16 words of each other. For example, "mysql NEAR odbc" - DataparkSearch will find any URLs that contain both "mysql" and "odbc" within 16 words of each other.

ANYWORD or * - ANYWORD operator, identical to AND operator, but come true if both words have any one word between and left operand have lesser position than right operand. For example, "mysql * odbc" - DataparkSearch will find any URLs that contain both "mysql" and "odbc" within any word between, for example, any document with "mysql via odbc" phrase.

OR or | - logical OR. For example, "mysql | odbc" or "mysql OR odbc" - DataparkSearch will find any URLs that contain word "mysql" or word "odbc".

NOT or ~ - logical NOT. For example, "mysql & ~ odbc" or "mysql AND NOT odbc" - DataparkSearch will find URLs that contain word "mysql" and do not contain word "odbc" at the same time. Note that ~ just excludes given word from results. Query "~ odbc" will find nothing!

() - group command to compose more complex queries. For example "(mysql | msql) & ~ postgres". Query language is simple and powerful at the same time. Just consider query as usual boolean expression.

8.1.8. The Verity Query Language, VQL

Only the prefix variant of the Verity Query Language is supported by DataparkSearch.

Also, only the following subset of VQL operators is supported by DataparkSearch:

Table 8-2. VQL operators supported by DataparkSearch

<ACCRUE>equal to OR operator in boolean mode.
<AND>equal to AND operator in boolean mode.
<ANY>equal to OR operator in boolean mode.
<NEAR>equal to NEAR operator in boolean mode.
<NOT>equal to NOT operator in boolean mode.
<OR>equal to OR operator in boolean mode.
<PHRASE>equal to a phrase in boolean mode.
<WORD>is considered as an empty operator.

8.1.9. How search handles expired documents

Expired documents are still searchable with their old content.