Brilliant query system

[0002] This application is a continuation-in-part of our application Ser. No. 09/630,227 filed Aug. 1, 2000.

[0003] The invention relates to providing an improved system for conducting queries over the Internet, (the Net), and private Intranets.

[0004] The Internet delivers trillions of words to billions of screens. The Net contains an enormous amount of material. Often when conducting searches on the Net one receives far too many search results, most of which are not relevant. One of the simplest ways to obtain more focused searches is to use a Boolean ANDed search with search engines such as, Yahoo, Excite, Google and Alta Vista. Instead of using one word, two words are used and much irrelevant material is discarded.

[0005] For example, searching on the word blackbird might yield tens of thousands of hits, with results ranging from rock bands, to birds, to consulting firms, to airplanes. Search on blackbird AND reconnaissance and the number of hits can be reduced by two orders of magnitude (from 6,000 to 60 hits) and almost all of the referenced Web sites (the “hits”) deal with the famous SR-71 spy plane (known as the “Blackbird”). The effect of the ANDed search is that both the words Blackbird and reconnaissance MUST occur in the verbiage of the Web page, and that the word blackbird is more important than reconnaissance because it comes first in the search, hence pages with more occurrances of the word blackbird will be sorted first in the resulting list by most search engines. Each of the search engines prioritizes the hits they return, on the basis of their own set of rules for importance, credibility or popularity. What these search engines do not do is to figure out just what you are really interested in.

[0006] It is an object of this invention to enhance a body of text to add focused and selected queries to the text.

[0007] It is a further object of the invention to provide a system to automatically add highly relevant and focused queries to a text, such as magazine articles, news stories or any other text.

[0008] Brilliant queries require a preparation process that analyzes any text to enhance and generate a set of suggested searches based on that analysis and certain pre-set user parameters. The output of this preparation process can be used to add links to an HTML page of a document either automatically or through manual insertion of the resulting analysis.

[0009] Brilliant query links have two components: a “hook” and a “keyword”. The hook is an overall concept or phrase that describes the subject matter of the text body. The keyword is a word that is derived from the analysis of the text and indicates a secondary or related concept. Brilliant queries are a collection of one or more pairings of the hook and a keyword. For example, an article on the SR-71 Blackbird Airplane might have the following brilliant queries:

[0010] 1. Search for more information on BLACKBIRD and AVIATION

[0011] 2. Search for more information on BLACKBIRD and ELINT

[0012] 3. Search for more information on BLACKBIRD and RECONNAISSANCE

[0013] 4. Search for more information on BLACKBIRD and TRANSPORT

[0014] The hook is BLACKBIRD and the keywords are AVIATION, ELINT (electronic intelligence) RECONNAISSANCE and TRANSPORT.

[0015] The hook is the concept, primary subject matter or main topic for a body of text. The hook is used to define a query as narrowly as possible on a particular topic for a selected information source. To determine a “hook”, a content layer must exist for which a context can be determined. There must be a perceivable structure to the information source and each content entry must have an associated context or place or places within the structure of the information source.

[0016] Information source is defined as any collection of content that is searchable for the purpose of locating specific content selections. Included in this are encyclopedias, news archives, dictionaries and other specific content collections. This includes search engines on the Internet that effectively turn the entire Internet into a single searchable information source.

[0017] Keywords are simply a collection of words, generated automatically or manually, that are deemed to be indicative of the topic matter or one of the topics for a given content selection. Keywords are determined by comparison of a predetermined list of keywords to the text of the content selection. If the content selection contains one or more of the keywords, then that keyword is associated with that text body and potentially used for the brilliant query. Keywords may also be determined by statistical frequency analysis of the text, with or without manual selection and addition of synonyms.

[0018] Stopwords are a collection of words that are used so frequently in a language that they provide no benefit at all as a search target for the selection of relevant content. This includes articles, conjunctions, prepositions, pronouns, etc.

[0019] Selection of Keywords

[0020] A brilliant query requires a list of keywords that are generated by automatic or manual statistical and empirical analysis of the body of content to be enhanced or a comparable body of content. The keyword list for a given content source is generated through the use of frequency analysis, stopword removal and finally, manual selection using empirical testing of the results generated by a given potential keyword. Based on experience, a solid keyword list usually runs between 250 and 1000 words and phrases which are chosen by the system designer.

[0021] Also, keywords can be manually tuned through the use of a thesaurus feature whereby a given keyword can be associated with one or more synonyms that would indicate the use of the keyword whenever one or more of the synonyms appear in the body of text to be enhanced.

[0022] Automatic Generation of the Hook

[0023] One embodiment of the brilliant query to enable an automatic process for generating brilliant queries for a body of text, is to determine the hook by extracting the highest frequency proper names from the text body. This process requires a two-pass analysis of the body of text. The first pass simply generates a frequency table with an entry for each word with the exclusion of stop words.

[0024] The second pass relies on the identification of proper names and punctuation to select a hook. Proper names are identified by locating all adjacent capitalized words not separated by punctuation. Frequencies for each proper name sequence are calculated by averaging the individual word frequencies by the number of words in the sequence. The hook is then selected by using the most frequently mentioned proper name sequence with the highest frequency. Also, if a word appears in multiple sequences, the longer sequence is given preference, even if it is a lower frequency than the shorter sequence.

[0025] For Example

[0026] “Governor Bush had a strong, substantive week,” Communications Director Karen Hughes said of a six-day, nine state swing in which Bush recovered from the verbal gaffes and tactical blunders that plagued his campaign in late August and early September.

[0027] “Governor Bush”—frequency (1.5);

[0028] “Communications Director Karen Hughes”—frequency (1.0);

[0029] “Bush”—frequency (2.0);

[0030] “August”—stopworded;

[0031] “September”—stopworded.

[0032] The analysis of the previous text results in “Governor Bush” being selected as the hook since “Bush” appears as a single word and as part of a phrase, the longer phrase is used even though it has a lower frequency than “Governor Bush”.

[0033] Common proper names such as days of the week and months are included in the stopword list. The automatic hook generation technique described here works very well for encyclopedic and news related content sources.

[0034] Automatic Generation of the Keywords

[0035] A word frequency analysis is done on all of the text, with stopwords excluded, and the resulting words, by order of frequency are compared to a pre-selected keyword list. Those that match, based upon a desired frequency become keywords to be combined with the hook to form focused, optimal queries.

[0036] Generation of Brilliant Queries

[0037] Once the keywords have been selected and the hook for a body of text has been determined or automatically generated, the searches are created by generating a link for every keyword extracted from the body of text and combining it with the hook in a search that results in a result set that is the logical intersection of the results generated by the hook and the keyword. Basically, each entry in the list of search results must contain both the hook and the keyword and not just one or the other.

[0038] This requires the use of special syntax for most search engines similar to the following:

[0039] Search Query:+“Governor Bush”+campaign

[0040] The + is commonly used to indicated a required word or phrase in a query for a search engine and the quotes indicate that the phrase “Governor Bush” is to be treated as a phrase and not as individual words. So the query above instructs a search engine to locate all articles containing the phrase “Governor Bush” and the word “campaign”.

[0041] After the hook and the keywords have been established the query is conducted selecting one of the hook-keyword sets. The selected set is then automatically converted to the appropriate syntax for a selected target search engine, to provide the best and most suitable query to obtain the most relevant query results, which must contain both the hook and the keyword or phrase.

[0042] FIG. 1 is a depiction of a computer screen showing a body of text;

[0043] FIG. 2 is a depiction of a computer screen showing the statistical frequency analysis of the hook;

[0044] FIG. 3 is a depiction of a computer screen showing the statistical frequency analysis of keywords;

[0045] FIG. 4 is a partial pre-selected keyword list;

[0046] FIG. 5 is a depiction of a computer screen showing selected focused queries.

[0047] The creation of brilliant queries consists of four steps:

[0048] 1. Selection of a corpus, that is, a body of text such a news report;

[0049] 2. Statistical frequency analysis of the word occurrence in the corpus;

[0050] 3. Editorial judgment or filtering of the keyword list;

[0051] 4. Application and tuning of the keyword set to target databases.

[0052] FIG. 1 is a body of text comprising a WABC Eyewitness News Online article entitled, “Lazio accuses Clinton of flip-flopping on Jerusalem”. Automatic statistical frequency analysis of the capitalized words in the article are shown in FIG. 2. Editorial judgment resulted in “Hillary Clinton” as the hook, the primary subject matter of the article, as the only person's proper name in the list. Automatic statistical frequency analysis of the article for keywords resulted in a number of keywords shown in FIG. 3. These included time, Palestine, peace, embassy, Israel, Jewish, America, union, politics, book and president. Those keywords can be matched to a pre-selected list of universal keywords, a partial list of which are shown in FIG. 4. Each of these combinations, hook (Hillary Clinton) and keyword (FIG. 3), appear on the screen (FIG. 5) as suggested focused queries available to the viewer to conduct a further search for the focused information in which he or she is interested. Any of the suggested queries may be selected and sent to a pre-selected search engine or information source, or may be sent to a search engine or information source selected by the user. When the search engine or information source is selected by the user, the system will place the query in the proper syntax for that particular search engine or information source. By using these focused searches, the user can retrieve only that information in which he or she is really interested instead of wide scattered, irrelevant results.

[0053] Word frequency studies of text is a well known process for the study of language. For instance, see a book entitled “Frequency Analysis of English Usage” by W. N. Francis and H. Kucera (1982) Boston: Houghton-Mifflin. See also, “Sum: Am. English Word Frequency Lists”, by L. Hillman. Applicants have used frequency analysis to obtain much more focused search results on the Net and other search sources.

[0054] This system is applicable to any body of text. Text that has a specialized subject matter, such as a medical text or a legal text, will be analyzed to create keywords that are associated with that subject matter. This system can be applied to any subject matter of any text.