---

Getting Started

The Evri Search system supports both simple document keyword search, and Relationship Search; a new means of searching for data that allows users to search for information conforming to a specific format. To use Relationship Search, users must specify a query that identifies what kinds of results they wish to see, based on actions that are performed or relationships between entities in the data.
Back to Table of Contents

---

Results Display

The power of the Evri Search system lies in its ability to summarize the found relationship in a concise and abbreviated representation. While document keyword searches return results displayed in a traditional list format, Relationship Search results are displayed in different format designed to help you find specific information within lots of results.

Each row is a single result. The first column contains the source of each relationship, or the entities that are performing some action. The second column contains the action (or verb) that defines the relationship, and the third column contains the target of the relationships, the receiver or object of the action.

The essential information from the sentence is extracted, summarized, displayed in table format. This allows you to see what's important in the results, without making you read a lot of text.

Note: results do not always have both a source and a target. In some cases, either the source or target may be left out, depending upon how the relationship was grammatically expressed in the text. In these cases, a dash ('-') is displayed in the appropriate column.

Back to Table of Contents

---

About Document Search

The Evri Search system supports document keyword searches by specifying a keyword or phrase, similar to other search engines. Simply enter your search terms in the text field and press the go button, and the results are listed as links to relevant documents.

In addition to terms and phrases, the following are also supported:

• Boolean searches using the operators "AND", "OR", and "NOT".

• Parenthetical expressions such as (X AND Y) NOT Z, where X, Y, and Z are search terms.

• Entity types or facets like [person], [musician], or [politician].

For document searches, any synonyms will automatically be searched. For example, if a user searches on U.S.A, other terms like United States will appear in the search results. If the user is only interested in the word or phrase exactly as it is specified, the word or phrase should be put in double quotes.

The Evri Search system also supports Relationship Search as an alternative to document keyword searches. Relationship Search is designed to help return more precise results in a format that is easy to scan quickly.

Back to Table of Contents

About Relationship Search

Relationship Search uses natural language processing to return very specific results in table format.

Notice that while a document keyword search may get thousands of results, Relationship Search typically returns a smaller number of very specific results for the same terms. Each row is a result that links to a single sentence where all these terms are connected, and all the documents where these terms are found in different places are filtered out.

The Relationship Search result display is designed to solve the problem of finding very specific data within large result sets, where desired information may not be found on the first few pages. One advantage of Relationship Search is that it allows you to define exactly what you are looking for, quickly scan through results, and go straight to the sentences that contain the information you seek.

Back to Table of Contents

---

Relationship Search Tips

When executing a document keyword search, the Evri Search system may recommend several Relationship Searches which may be of interest based on the keywords you chose. These recommendations are presented as links in the right "related to" column and execute a Relationship Search.

For example, if you are searching on a certain topic, you may see a series of links presented to the right of the result set that will execute Relationship Searches that link your keywords with people, places, or other related topics. You can execute one of several Relationship Searches by clicking on these links.

When you click on the link, instead of seeing a list of document results, you will see a table display containing the Relationship Search results. Information about interpreting Relationship Search results can be found in the Displaying Results section.

Back to Table of Contents

---

Using Relationship Search

This section contains information about using Relationship Search and how your search results can be improved by creating custom queries.

The first three topics in this section, How to use Relationship Search, Constraining your search, and Searching with actions, discuss how to create good queries.

The other topics will be more relevant to users who wish to compose their own Relationship Search queries from scratch in order to exploit more powerful aspects of the Evri Query Language.

Relationship Search queries are based on the Evri Query Language. At its most basic level, the query language is based on the format:

Source > Action > Target

To create a query, you need to specify either a source, or an action, or a target, or any combination of the three. The source can be any person, place, or thing, like Madonna, performer, or London. The action can be any verb expressing an action, like sing, perform, or visit. The target is who or what the action is directed towards.

How to interpret these results is explained in detail in the Results Display section.

To learn more about how to improve your Relationship Search query, see Constraining your search.

Back to Table of Contents

---

Constraining Relationship Search

Relationship Searches can sometimes produce too many results, or results that are not relevant. In such cases, changing the terms in your query is one alternative. Another alternative that works well is to specify one or more additional keywords that can be used to constrain the search. You can also improve your Relationship Search by Using Taxonomical Paths.

Back to Table of Contents

---

Relationship Searching with Actions

In order to write effective Relationship Search queries, it is important to understand how actions are processed.

Tenses

Actions are defined by verbs or groups of verbs. In general, verbs should be specified in infinitive form or in present tense. If the verb in the query is not in present tense, it is normalized to its infinitive form. If it is double-quoted and not in infinitive form, it will not be normalized and it is less likely that results will be returned. If the verb is in present tense, by default all forms and tenses of the verbs will be included in searches. For example, if the query includes the verb talk, results will also include relationships that contain the forms talked or talking.

Similar verbs

If a verb like talk is specified, similar verbs like speak will also be searched in all its various tenses. The Evri Search system maintains a list of similar verbs that are included in Relationship Searches (but not Document Searches) by default.

If users specifically wish to search only on the verb in the query and no other synonyms, verbs can be quoted, as in:

Oprah > "promote" > Obama

In this case only the verb "promote" will be searched.

Back to Table of Contents

---

Creating Relationship Search Queries

This topic assumes familiarity with Results Display and Using taxonomical paths.

Entities and actions: building blocks of a Relationship Search query

A query is made up of Entities and Actions that are linked via a series of operators. At its most basic level, the query language is based on the format:

Source entity > Action > Target entity

To create a query, you need to specify either a source, or an action, or a target, or any combination of the three. Let's define them in more detail:

• Entity - An Entity is any noun or noun phrase in the search query or result. An Entity can be the source (initiator of an action), the target (receiver of an action), or the complement of a prepositional phrase. Entities can be multiple words. If they are quoted, the exact phrase must be matched by a phrase in a document being searched. Either double quotes or single quotes may be used; if double quotes are used, then synonyms of the quoted expression will not be included in a search. If single quotes are used, synonyms of the quoted expression will be included. (Note that entities cannot start or end with a dash ('-') unless quoted).

• Source - The initiator of an action is referred to as the source. For example, in the query [Politician] > criticize > John McCain, "Politician" is the source. Here we are interested in all politicians that are criticizing John McCain, but not all politicians that John McCain criticizes.

• Target - The receiver of an action is referred to as the target. For example, in the query  [Actor] >star> [Movie], "Movie" is the target of the action. Here we are interested in all actors (and actresses) that are starring in films, but not all films that are starring in actors.

• Prepositional Complement - An action is often performed with a prepositional complement. For example, in the query [Athlete] > score > points PREP CONTAINS 2007 , "2007" is the prepositional complement of the sentence. We are only interested in point scorings that happened in 2007.

• Action - All relationships are based on an action, or verb. For example, in the query [Athlete] > score > points , "score" is the action.

 

When writing a query by hand, entities and actions are combined using the directionality arrows <, >, <>. These arrows indicate the direction of the action taking place. For example,

[Person/Name] > criticize > McCain

... will return information about people that criticize McCain of something. Conversely,

[Person/Name] < criticize < McCain

... will return information about people that McCain is criticizing. (Note that the arrows have changed direction.) If we specify arrows in both directions, like this:

[Person/Name] <> criticize<>McCain

... we will get information about both McCain criticizing and being criticized.

Using boolean operators

Relationship Search supports the use of boolean operators, as shown in the example earlier. Note that the boolean operators can be nested, as in:

england AND NOT (aerospace OR airways) > abandon > *

The query can also be restated as:

england AND NOT aerospace AND NOT airways > abandon > *

The default operation for omitted boolean operators is OR. Booleans do not have to be uppercase, although they are presented that way here for clarity.

Noun phrases such as "New York" can be included in quotes, as in:

Bush OR "New York" > * > *

Parenthesis can also be used, like:

"New York" OR (Paris AND London) > * > *

 

Wild card characters

An asterisk (*) can be used to denote unknown or unspecified sources or targets. For example,

Brad Pitt > * > *

... is a valid query that will return references to Brad Pitt in documents where he is the source of any action, directed at anyone or anything. The opposite query:

* > * > Brad Pitt

... will return references to Brad Pitt in documents where he is the target of any action, directed at him by anyone or anything. Alternatively, you can include the '*' wild card character as part of a term, as in:

Angelina Jol* > say > Brad Pitt

... which will return all relationships where an entity starting with "Angelina Jol" (such as Angelina Jolie) said something to Brad Pitt. You can also use the '?' wild card character to indicate only a single letter to be matched. For example:

Bin Lad?n > meet > [Person]

... this is useful, because "Bin Laden" is sometimes spelled "Bin Ladin".

Constraining your query

We call the source > action > target component of a Relationship Search query the relationship. In addition to the relationship component of the Relationship Search query, there are three optional clauses that can be added to filter results:

• any prepositional constraints, to filter results by information found in a prepositional phrase;

• any document keyword or taxonomical path constraints, to restrict search to documents that have certain keywords or taxonomical paths;

• any relationship context constraints, to restrict search to relationships where certain keywords or taxonomical paths are found within one sentence of the sentence containing the matching relationship.

Prepositional constraints add a further level of specificity. In addition to searching for a source > action > target relationship, you can specify that another keyword should be included as part of a prepositional phrase. For example, if you are searching for corporate acquisitions in the United Kingdom, you can constrain your query like this:

[Organization/Name] > acquire > [Organization/Name] PREP CONTAINS United Kingdom

This will return only results that reference sentences that have prepositional phrases with United Kingdom in them, such as in the United Kingdom. You can also use taxonomical paths in a prep constraint, such as:

[Organization/Name] > acquire > [Organization/Name] PREP CONTAINS [Money]

[Money] is a taxonomical path that includes any monetary amount. This query would return information about all corporate acquisitions only if there are monetary expressions mentioned, like for 2 billion dollars , for 500K EUR, or about 3M Rupees.

Note that a context constraint with the same term will return more results than a prepositional constraint, and in general, the grammatical subject or object results will rank higher than other context matches, such as a result in a previous or subsequent sentence. For example, the query above will return more results, if specified like:

[Organization/Name] > acquire > [Organization/Name] CONTEXT CONTAINS [Money]

Combining clauses

These clauses described above, if combined, must appear in that order, and must be separated by at least one white space.

These clauses can be expressed in either a long or abbreviated format. In the long format, the clauses are separated by the self-explanatory terms "PREP CONTAINS", "DOCUMENT CONTAINS" and "CONTEXT CONTAINS". Look at this example, broken up into several lines for easier reading:

Bush > visit > [Country] AND NOT China

PREP CONTAINS [Date]

CONTEXT CONTAINS "foreign service" OR diplomat

Here we see a Relationship Search query that specifies a search for visit relationships between the entity Bush and any country except China. The Relationship Search query is constrained by the preposition [Date], meaning that a date must be included in a prepositional phrase within this relationship. The search is further constrained by the document context keywords / keyphrases foreign service and diplomat, meaning that only relationships from documents containing these words within one sentence of the relationship should be returned.

Put together, this represents a powerful query that will search specifically for diplomatic trips that Bush took by plane to foreign countries with the exception of China on a particular date. Note that although the query is separated into three lines here for clarity, it is interpreted as a single string by the Evri Search system. Of course, queries need not be so specific or constrained; the simpler queries shown above that do not contain document or metadata constraints will simply return more results.

Here we have specified two expressions for the document filter: foreign service and diplomat. In addition, if a document contained the adjective form of the diplomat, that is, the word diplomatic it is also included. The search system automatically extracts the stem of the word and searches for other forms. Sometimes when you perform searches, you will see that your query has been "stemmed" or truncated to remove a final 's', 'ed', or other non-essential parts of the word. Such changes to your query are presented in green text so that this will be clear to you.

For convenience, constraints can also be entered in a more abbreviated form. The terms "PREP CONTAINS" and "CONTEXT CONTAINS" are replaced by a '^' and a '~' character respectively. In our example, this would look like:

Bush > visit > [Country] AND NOT China ^ [Date] ~ "foreign service" OR diplomat

This query is processed in exactly the same way as the one above.

Using noun phrases and modifiers

Within the Relationship Search query, the sources or targets of an action can be either nouns or noun phrases, like "No Country for Old Men". Sometimes many different noun phrases describe the same things, like "prostate cancer" and "cancer of the prostate". Because modifiers of key nouns are also searched by the system, you should be able to find all results you are looking for even if they are expressed in different ways. Similarly, you could find all actions involving an organization, like the National Transportation Safety Board, regardless of whether it is referenced by its full name or simply as "National Transportation Board".

Special characters

Certain special characters may not be interpreted by the system correctly, and should be avoided if possible. The current list of special characters is the following:

+ - & | ! ( ) { } [ ] ^ " : \

Term Offsets

The Evri Search system supports term offsets, in which users can specify that two or more keywords must be found within a given number of words of each other. For example, the following query:

 "malignant cancer"~10  

would return all instances in the corpus in which the words "malignant" and "cancer" are found within 10 words of each other. This allows users to search for specific terms that may be separated in the documents by several other words, or several lines of text.

Also note that the symbol '~' is used for both term offsets and for the CONTEXT CONTAINS clause. When used after a keyword and followed by a number this symbol is interpreted as the term offset operator.

Sample queries

Here are some more example queries that express some of what has been discussed so far. Note that if you execute these, you may see the short form of the query (with '~' ) in the search field rather than the longer form displayed here (CONTEXT CONTAINS).

Back to Table of Contents

---

Searching With Taxonomical Paths

A powerful feature of the Evri Search system is its support for specifying taxonomical information within a search query to search for categories of information.

Using Taxonomical Paths

Another way to improve your Relationship Search is by specifying a taxonomical path instead of a specific word or phrase for your source or target. For example, what if you wanted to know all actors or actresses who have entered rehab? With most search engines, you may have to make a list of all the possible or actors or actresses and search on each of them individually, because there is no way to search for "any actor that..." or "all actors that...".

Evri Search offers a way to do this, using taxonomical paths. Taxonomical paths express type or class information that can be used to allow users to search for specific types of entities, like 'cities' or 'radio personalities'. If you are interested in searching for radio personalities, you could simply enter in [Radio_Personality] as a search term, which is interpreted as "any radio personality".

Taxonomical paths are also useful in Relationship Searches. For example, if you are interested in searching for information about artist drawings, you can simply specify the query

[Artist] > draw

Note that taxonomical paths are hierarchical. The Evri Search system recognizes "city" as within the hierarchical path "Location/City". This means that if a term in the corpus is mapped to "city", it is also understood to be a "location". Therefore if locations are searched for, any cities and other sub-paths are returned along with terms that are specifically mapped to "location".

Back to Table of Contents

---

Evri Search Taxonomy

The Evri Search system supports the following standard taxonomical paths. You can use these taxonomical paths to search for information about types of things. For example, you can use [person] in your Relationship Search query to return results about people in general, like:

John McCain > meet > [Person]

This query would return results containing all sentences found in which John McCain met with somebody. Note that to use a taxonomical path, you do not need to specify the entire path, only the last term. Some terms may refer to more than one taxonomical path. For example, [Name] refers to people (Person/Name), locations (Location/Name) and organization names (Organization/Name).

For more information about using taxonomical paths, see the Using taxonomical paths section.

Back to Table of Contents

Query Language Cheat Sheet

The following table demonstrates the basic kinds of queries that Relationship Search supports. This page assumes some familiarity with Relationship Search. Note the use of '*' as a wild card character, meaning, "any action performed" or "any entity found".

Back to Table of Contents