|
|
This appendix describes the default behavior of the portal searches. This information is available to users through online help. This appendix includes the following sections:
The portal provides basic and advanced search tools for typical and advanced users, respectively. The fundamental search syntax and behavior are the same in basic and advanced search, but basic search adds automatic broadening, ranking features, and syntax correction. The following table specifies the search type implemented in the search tools available through different areas of the portal.
This section describes the expected behavior of supported search syntax. It includes the following topics:
The Search Service parses queries to determine which of the following operator modes to use for the query:
Query operators AND, OR, NOT, and NEAR are spotted without any special marking (for example, cat AND dog), but all other operators must be surrounded by angle brackets (for example, <WORD>) to be recognized as having special meaning.
A query that contains three or more terms and an operator is parsed as if the terms on each side of the operator were quoted phrases. For example:
Search Service and Notification
"Search Service" AND Notification
Search operators are localized for the following European languages: English, Danish, Dutch, Finnish, French, German, Italian, Norwegian (Bokmal), Norwegian (Nynorsk), Portuguese, and Spanish. If you put angle brackets around the operators, the English versions are also recognized. For example, in the Spanish locale, the following queries are equivalent: perro Y gato, perro <AND> gato, and perro gato. However, perro AND gato is not equivalent in the Spanish locale, because AND is not surrounded by angle brackets.
Anything enclosed in angle brackets but not recognized as one of the supported operators is ignored.
+dog -cat bird
This query returns documents that contain dog but do not contain cat, and ranks documents with both dog and bird highest. Compare this to a similar query:
bird -cat
This query returns documents that contain bird but do not contain cat. Absent any + terms, the plain term bird is treated as a required term.
The following table summarizes the behavior of operators.
There are certain circumstances in which a user can unintentionally invoke a more advanced search mode by inadvertently using operators. Examples include the following queries:
In each of these examples, enclosing the query in double quotes yields the desired effect.
The Internet Style mode operators '+' and '-' take precedence over the other search operators. For example, +big dog <order> cat matches all documents that contain the term big, boosting the ranking of any documents that contain any of the three terms dog, or cat.
Within query operators mode, the operators have the following precedence classes, from greatest to least:
Parentheses can be used to override operator precedence. The following two queries are equivalent (the parentheses do not effect the semantics of the search).
This search matches documents that meet one of two conditions:
On the other hand, the parentheses in the following query override the default operator precedence:
a and b near (c or d)This search matches documents containing the terms a and b and either c or d, where b is in close proximity to c or d.
Punctuation is treated specially. The following rules describe the interpretation of punctuation characters.
HOST_NAME, but without the quotes, it also matches HOST NAME). Symmetrical punctuation tokenization takes place on text stored in the index, so the explosion of a query term such as bag-of-words does not prevent the search from matching a document containing the phrase bag-of-words.
All searches are case-insensitive, except when the <WORD> operator is used.
Word stemming is applied to all individual terms in the search query, except within quoted phrases, or when the <WORD> operator is used. The stemming of query terms means that a query term will match documents containing morphological variants of that term. For example, a search for dogs AND go would match a document containing the terms dog and went. (This example applies to English; stemming employs language-specific information and depends on the user's locale and the language used to index the document.)
The wildcard operator is used to search for prefixes, suffixes, and substrings of indexed terms. Wildcards cannot be used within quoted phrases.
Terms generated by wildcard expansion are not stemmed.
Wildcard expansion is performed internally by replacing each pattern with a limited list of terms that match the pattern before actually executing the query. Very broad wildcard expressions might therefore return a partial list of results.
A quoted phrase in the user search query matches only documents that contain the given sequence of terms. For instance, a search for "big dog" will not match a document that contains the terms big and dog if it does not contain the phrase big dog. Stemming is not applied to terms within a quoted phrase. Also, wildcards cannot be used within quoted phrases.
If thesaurus expansion is enabled, then thesaurus expansion is applied to the individual terms in a basic search. Thesaurus expansion is applied in all three search modes (Internet Style, Query Operators, and Bag of Words). Thesaurus expansion is not applied to quoted phrases. If a term is expanded by a thesaurus entry, then it is not eligible for automatic spelling correction.
Unlike automatic spell correction, which is applied only as a fallback when the non-corrected terms do not match any documents, if thesaurus expansion is enabled, then it is always applied to all individual search terms.
Documents and portal objects are indexed with a language setting that determines how word breaking and stemming are applied. When a user issues a search query, word breaking and stemming are applied according to the user account locale settings. Search results are best when the language used for the search matches the language of the documents being searched. However, searches are normally applied to documents in all languages. Cross-language searches do not benefit from localized stemming and word breaking, but can still return useful results.
The advanced search page offers the ability to restrict searches to a particular language.
When searching portal content via the Search box in the portal banner, the text of the query is processed using the language setting of the user interface. If the portal interface is German, the query is tokenized and stemmed using German language rules, providing optimal search results for documents indexed using German linguistic rules.
If the search collection contains documents in other languages, you can still retrieve them with a query using the appropriate text (assuming the user interface permits entry of the necessary characters). Typing English words into the search box of a portal using a German interface applies German linguistic rules to the query text. Because English stemming is not used, the query is not able to match alternate English word forms; however, English language documents containing the entered words are retrieved.
Although you can enter Asian language text into a European language search box (if a compatible character encoding is used), you should limit the text to a single word or manually separate words with white space to be able to match Asian content in the search collection.
The Advanced Search page provides additional functionality for searching in a multi-language document collection. A pop-up list allows the user to select the language to use for query processing. Linguistic rules for tokenizing and stemming the selected language are used when processing the query text. Among other things, this means that Asian text can be entered without unnecessary white space.
The query operators recognized by Simple Search and Advanced Search are sensitive to the language setting. For example, the AND operator can be specified as "UND" when the query is processed as German. Localized operators are available for the following languages: English, Danish, Dutch, Finnish, French, German, Italian, Norwegian (Bokmal), Norwegian (Nynorsk), Portuguese, and Spanish. All other languages use English operators.
The descriptions of searches below do not include any of the query expansion or ranking techniques that are employed in basic search. Except where otherwise noted, all matches are case-insensitive.
Search results are ranked according to relevance, by default. The following topics in this section describe the factors that determine relevance:
The number of times a query term (or its stemmed and case variant forms) appears in a searchable item has a large influence on the relevance ranking of the item. All other things being equal, items which contain more instances of a query term will rank higher than items containing fewer instances. This is known as term frequency based ranking.
Basic searches are performed across several document fields, and some fields are weighted higher than other fields, so that, for instance, a match on an object name ranks higher than a match on an object description. By default, the fields searched are name, description, and full-text content. For information on modifying or adding field weights, see Modifying the Properties Searched and the Relevance Weight for Properties.
In basic search, Bag of Words mode employs special relevancy ranking features which emphasize phrase and proximity matches with the search phrase, even though the user did not employ quotes or proximity operators.
The search phrase terms are used to generate three queries:
The three queries combined with the OR operator into a single query, and the relevance ranking are designed to ensure that the results from group #1 always rank above group #2, which rank above group #3.
For example, if you enter "san francisco" hotels, the following queries would be generated:
The search results pages for banner and advanced search allow you to sort the search results by last-modified date, folder, or object type.
Basic search adds some special features in order to increase the chances that a search will return relevant results. As noted in the previous section, term proximity can boost the relevancy ranking in basic search. Automatic spelling correction is also applied only in basic search.
In basic search, if a user search query causes syntax errors in Internet Style mode or query operators mode, it is automatically retried in Bag of Words mode to be as forgiving as possible of user error. For example, if you enter dog and, this query would cause a syntax error in Query Operators mode, because it is missing the right-hand operand to and. The query would then be passed to Bag of Words mode, which would attach no special operator significance to and, and would therefore retrieve documents containing dog and and.
Advanced search behavior is intended to support complex, precise queries. Therefore it generally does not employ the automatic broadening features of basic search, such as broad cross-field searching or automatic spell correction. Stemming, however, is applied in advanced search.
The Text Search portion of advanced search will search across name, description and full text content. Additional property criteria are applied only to the fields specifically selected in each criterion.
User queries that cause syntax errors in Internet Style mode or Query Operators mode will display an error message in the user interface; the search will not fall back to Bag of Words mode.
|
