CA2610517A1 - Process for analysis and processing of requests applied to a search engine - Google Patents

Abstract

The system includes: a site (40) interfaced to a search engine; means (10, 20) for generating vocabulary words; means (42) for submitting to the engine queries each containing a word or group of words; and means for receiving a result representative of the occurrences of the word, or the association of words, in web pages indexed by the engine. The method includes: (a) determining a start word and submitting a query; (b) receipt of an initial result; (c) the generation of an additional word and the submission of a request; (d) receiving the result, lower than the initial result, corresponding to the request formulated in step c); (e) repeating N-1 times steps c) and d) with a different additional word, giving N respective results; and (f) the ranking of these N results.

Description

Method of analysis and treatment queries applied to a search engine Technical area The invention relates to a method for analyzing and processing requests applied to a search engine.

More specifically, it aims to improve the performance of these engines.
research of in a more efficient and economical way (in terms of resources the extraction of the information sought.

Prior art The recent development of applications related to the use of the Internet has made popular the use of search engines such as Google (trademark of Google Inc.) or Exalead (registered trademark of Exalead SA) for the search for information, thanks to to the ability of these search engines to index the content of several billion pages accessible to consultation on all kinds of websites.

These search engines (hereinafter "engines") are implemented in their addressing a a query containing a word or a "sentence", ie several words (in the absence particular operator such as AND, OR, etc. the words are supposed to be connected by an AND
logic).

The engine responds to this request in a fraction of a second, providing a numeric value (called "Results") representative of occurrences ("hits") of pages containing this word (or the words of the sentence) among all pages indexed by the search engine and a list of relevant Internet sites.

The number of occurrences (hereinafter "result") proposed in return by the engine depends greatly from the choice of words that constitute the request. For some words, he can reach tens or even hundreds of thousands or millions, while than for others it is reduced to a very small number (a few tens, see units). Yes the user submits

-2-very general terms, the result will be extremely high; if, at the opposite, it uses a very precise designation, the result will be much smaller.

The first generations of search engines operated by analyzing the pages indexed on a criterion of exact concordance, literal, of the word supplied with the request.

This led users to often trial and error multiplying queries, modifying the terms submitted to the search engine, adding or subtracting certain terms, etc. and this in a totally empirical way and without any certainty that the modifications made can help to really converge the engine towards the subject of the search. In addition, experience shows that users badly evaluate the discriminative power of words with regard to computer search engines in the research process specific to this one, leading to results erratic, badly mastered.

In addition to the risk of omitting relevant answers, the result is technical view unnecessary multiplication of the number of requests, with an increase correlative traffic on the network and, at the level of the engine, a significant increase in resources necessary to deal satisfactorily and in a timely manner reasonable the flow of queries.

However, the increasing diffusion of the Internet and the use of search engine by the general public, as well as the concentration of requests on a number of more and more reduced number of engines considered as the most efficient, suggest a considerable increase in computing resources that will be needed to these search engines in the coming years.

These computing resources concern both the computing power needed at processing of the request itself, that queue management, whose complexity and storage capacity needs increase according to the traffic incident.

Current developments are moving towards research no longer based on identity the exact word, or even variants of it (singular / plural, form declined, writing variant, etc.), but on related words, that is to say belonging to the same

-3-semantic field than the initial word, so as to increase the potential of research and allow faster convergence to the subject of the request.

Such an algorithm is proposed under the name of Google Sets (registered trademark of Google Inc.), which is a predictive software for generating words (names common or nouns), for example a list of fifteen words, starting from only two or three words given by a user, the list provided being deemed to belong to the same field semantics as the words given by the user. For example, the user type the name of two or three American states, and the generator produces a list of states of this country. he can in the same way generate names of presidents of the United States, of society, etc. from two or three examples given to him. The software knows well perfectly discriminate {Johnson Pfizer} who will give a list of pharmaceutical companies, of with {Johnson Washington} who will give the names of presidents of the United States.

WO-A-2004/031916 (Google Inc.) describes an algorithm of this type, for the selection of clusters of words in conceptual relation with one or more words constituting input parameters of the algorithm. Clusters are chosen from a model probabilistic complex, so as to characterize, from a semantic point of view, the document from which the initial words are extracted.

But this algorithm is essentially a generator of words or clusters of words related; it does not aim at assessing simple semantic proximity, for report to a word initial, of a given word from an external source.

Summary of the invention There is therefore a need for a tool capable of evaluating semantic relevance of a word given (hereinafter "additional word") or a family of words considered collectively, by relation to another given word (hereinafter "starting word") representative of the field general semantic object of research.

The invention proposes a tool that can be used with words (words of departure and / or additional words) produced by an operator or automatically by a generator of words, or chosen and submitted by a user. These words will constitute the settings

-4-input of the algorithm implemented by the method of the invention - which has not not for purpose to produce words, but to assess the relative relevance of the point of view semantics.
Another object of the invention is to propose such a method which enables to improve non-empirically the process of handling query queries applied to a search engine, with the following results in particular:
time to treatment and, correlatively, the acceleration of the final response to the user ; the reduction the overall number of requests, allowing optimization of the management of queues waiting and associated storage resources; an overall reduction traffic, including a better fluidity of it.

Moreover, besides these material and quantitative improvements directly attached to the search engine and network operation, the user will benefit incidentally of an improvement of the quality of research, the method of the invention being able to propose a new navigation heuristic replacing the method current, empirical and uncertain in its results.

These improvements will notably benefit people using bases of documentary data, in applications such as parts management spare, archival documents, etc. and in particular, according to a characteristic particular of the invention, by optimizing access to the most solicited.

Incidentally, the process of the invention may even be used for teaching or fun, for example to bring a user, confronted with a word of given departure, search for the most relevant additional words from the point of view of their distance semantics. The process will be used to return to the user, for each of the words submitted, a quantification of their relevance to the word departure. The method can also be implemented between several users, for example example for compare their respective abilities in search of relevant words.

The system includes: a site (40) interfaced to a search engine; of the means (10, 20) to produce vocabulary words; means (42) for submitting to engine queries each containing a word or a group of words; and means for receive a result representative of the occurrences of the word, or association of words, in web pages indexed by the engine. The method comprises: (a) determination

-5-a word of departure and the submission of a request; (b) the receipt of initial result; (vs) the generation of an additional word and the submission of a request; (d) the receipt of result, lower than the initial result, corresponding to the request made to step c); (e) the reiteration N-1 times of steps c) and d) with a different additional word, giving N
respective results; and (f) the ranking of these N results.

The computer system comprises: a site (40) interfaced to a search engine research ; of the means (10, 20) for generating vocabulary words; means (42) for submit the search engine queries each containing a word or a group of words;
and means for receiving in response a result representative of the occurrences of the word, or association of words, in a set of web pages indexed by the engine of research. The method comprises: (a) determining a start word and submitting a query containing that word; (b) the receipt of a result initial corresponding to this starting word, (c) the generation of an additional word and the submission a request including the word of departure and the additional word; (d) the receipt of the result, less than the initial result, corresponding to the request formulated at step vs) ; (e) the reiteration N-1 times of steps c) and d) with each time an additional word different, thus giving N respective results; and (f) the ranking of these N
results. After the stage f), there is further provided a step (g) of assignment, according to a rule according to N results, a global index of affiliation attributed to the family of N words additional generated, and relative, assigned in consideration of the starting word. This index can include be compared to a threshold, preferably a variable threshold, parameterizable by the user, whose crossing does or does not result in the production of an indicator Boolean.

Brief description of the figures Fig. 1 is a schematic representation, in blocks, of a system computer science capable of implementing the method of the invention.

Detailed description of an embodiment of the invention In FIG. 1, reference numeral 10 denotes computer terminals, with for each one a microcomputer 12 connected by an interface 14 to a network of telecommunication 16, which may be the wired telephone network (in switched mode or in ADSL mode), a network

-6-cable television, or an Internet connection via a server common to several positions.

The system may also include a number of terminals phone 20, connected by a microwave switching link to an interface 30 like mobile phone switch. The telephone terminals 20 are equipped with functions enabling them to exchange digital data, in particular textual, with the interface 30 according to various well-known technologies such as SMS, WAP, GPRS, UMTS, etc. The user enters on the keyboard data that will be sent to the interface 30, and receives from the latter messages that will be displayed on the screen of phone.

The system also includes a central computer site 40, typically a site Internet. Functionally, this site 40 comprises a unit 42 forming an engine of search (or coupled to a remote search engine), associated with a unit 44 from formatting, with two subunits 46 and 48 capable of formatting the messages in function the type of terminal used. Specifically, subunit 46 formats the messages so that they can be received and displayed by the microcomputers 12 by example under the form web pages viewable by a browser, then sends them to the network 16 over there 50. The subunit 48 instead performs a formatting adapted to a posting on mobile phones, for example in the form of WAP pages sent to the interface 30 via the link 52. The interface 30 is also connected to the unit 42 via a link 54 transferring data received from the terminals to the latter phone 20 connected to the network. It should be noted that the content of messages formatted by units 46 and 48, that is the content of the messages exchanged respectively with the computer terminals 10 and telephone 20, is identical; alone change the setting on page depending on whether the information should be displayed by one or the other type of terminal.

It should be noted that the invention can be implemented by other systems IT
than the one just presented, which has no limiting character.

In particular, the method of the invention can be implemented either by a dedicated device, integrated into a pre-existing application such as a browser or spreadsheet by a

-7-software module or macro-instructions running as a function specific the process steps that will be described below. Formulas and macro-instructions from a spreadsheet software such as Excel (trademark of Microsoft Inc.) animated by a macro-language QuicKeys (registered trademark of CE Software Inc.) are particularly suitable for execution of the various calculations and classifications set out below, as well as presentation results and the introduction by the user of the various parameters and data.

For the simplicity and clarity of the presentation, the following description will be made in the hypothesis where the different terms analyzed are words chosen by people (users), who enter these words for example on the keyboard of their computer or their mobile phone for transmission to the central site, and receive from the central results treatment, which they themselves exploit for their own purposes.
But the invention is equally applicable to configurations where the terms are produced by word generating algorithms, as in the case of WO-A-2004/031916 above, which can automatically deliver clusters of conceptually related words by application of a probabilistic model.

Similarly, the invention is applicable to situations where the results delivered speak method of the invention are reused in an automatic processing, by example for optimize the search for groups of words belonging to the same field semantics, to compare the relative relevance of various groups of words, to improve the convergence research, etc.

In other words, the method of the invention can be implemented, in any case or part, with a "user" who is not necessarily a natural person, which means this process can be interfaced both upstream (introduction of parameters) downstream (exploitation of the results) with a separate device ensuring a treatment automatic all or part of the data.

In general, the starting point of the invention lies in the finding out what a engine not only provides referrals to all pages containing a given word, but also provides other information, itself of great value, to know the number of pages where the word appeared. This information is the one that appears usually under the heading "Results" returned by the engines.

-8-For example {capacitor} gives a result of 1,140,000 occurrences (by convention the braces used in the text of this description indicate the term, or the series of terms, submitted to the engine). This result is better than that of {Multiplexer}
(188,000), but much lower than {resistance} (2,020,000).

A finer function of the engines is to look for non more than one word unique, but two, three or more words together in a "sentence" in the even request. For example {resistance inductance} returns 104,000 occurrences. From where one can deduce that among the 2,020,000 pages that contained {resistance}, only about one in twenty also contained the word {inductance}.

It is this property of the motors that the process of the present invention is specifically exploit.

Essentially, the basic idea of the process of the invention consists, after have selected a first word (the "starting word"), to find another word (the "word"
additional ") which will produce the highest result.

Thus, the starting word {transport} reveals about 48,000,000 occurrences. he is therefore for the user to search for an additional word that, combined with {transport} will produce the number of occurrences as high as possible. For example, {coconut palm} is a bad selection: the engine finds only about 2,300 occurrences for the phrase {transport coconut tree}.

-9 While {plane}, {metro} and especially {train} provide more results high, such as shown in Table 1:

Words Result transport 48 million aircraft transport 585 000 subway transport 1,250,000 transport train 2,800,000 transportation travel 1,140,000 car transport 611 000 Table 1. Results for some additional words combined with {transport}

We observe that it is {train} the "best" word, in that its result is superior to those from {metro} and {car}.

But by looking a little more in what he considers to be the lexical field of {transport}
the user discovers that it is possible to do better than {train}, such as indicated in the table 2:

Words Result Bus transport 3 400 000 Table 2. Results for word {bus} combined with {transport}

Or for {bus} a result superior to that of {train} It is thus finally {bus} which is the "best" word - at least for these six attempts.

In the case of an additional word consisting of a single term of a word compound in the form of a combination of terms, for example {"path of iron "} in the example above (quotation marks between braces corresponding here to the syntax used to query the engines, ie the engine will search for occurrences of the inseparable string of words enclosed in quotation marks).

-10-Advantageously, the method provides, before providing the result of the sentence containing the additional word selected, to check the result of the additional word considered in isolation, and to refuse this additional word if it presents, alone, a result superior to the word departure. This avoids the process being biased by the selection as a word addition of an insignificant term such as an article or an adverb, etc., who would give (often) a high result.

The course of the process can be expressed in particular by the flowchart simplified in metalanguage of Table 3, which is given for illustrative purposes only (MD stands for the word of departure, MA the additional word).

The result can be presented either in number of occurrences of the combination {(word of (additional word)}, or in the form of a ratio between the latter result and result the starting word taken in isolation.

input (MD) get result (MD) counter = 0 while not STOP

input (MA) test result (MA) <result (MD) get result (MD & MA) if result (MA)> Highest then Highest = result (MA) counter = counter + 1 if counter> 10 then END
wend Table 3. Flow of the metalanguage process

- 11 -So, in the following example of Table 4 corresponding to queries in the domain key components of electronic engineering (the aforementioned ratio being multiplied by 100 to improve readability):

Word (s) Result Ratio * 100 engineering 1,100,000 100 engineering memory 595,000 54,0909 engineering register 159,000 14,4545 engineering relay 150 000 13.6364 engineering meter 71 100 6,4636 engineering buffer 54 700 4.9727 engineering integrator 54 300 4,9364 engineering decoder 44 200 4.0182 engineering shielding 41 400 3.7636 engineering multiplexer 15 500 1.4091 engineering schottky 936 0.0851 Table 4. Example of requests in the field of electronic engineering For a ratio less than 10/00 (result less than 0.1), the additional word is considered as "irrelevant", irrelevant, which will not be considered as belonging at semantic field of the starting word.

This property can in particular be used in a variant aimed at search for words whose relative semantic distance is the highest, for example for isolate certain words that we want to establish with certainty - and in an automated way -that their probability is very small to belong to the same semantic field as the word departure. In other words, the process is no longer based, in this case, on the "meaning" of words but at contrary on the "misinterpretation".

-12-In yet another variant, instead of initiating the process with a word of departure consisting of an isolated term (eg {engineering}), one can take as word of starting a combination of words (eg {memory engineering}).

It is thus possible, after a first execution of the process started with a word from isolated departure, to retain one of the combinations (word-of-departure, word additional) which has been submitted to the search engine, and rerun the algorithm with this combination of particular words as new starting word.

A weighting may be applied to take into account the number of attempts, by applying a reduction which is all the greater as the number of attempts is Student. Conversely, a bonus may be applied if the "best"
word is found at the first attempt.

In a particular form of implementation, for a given starting word (for example {transport}) the central site selects a plurality of words (e.g.
{bus}, {train} and {metro}) and presents these words to the user, without indication of result and in one order anybody, asking him to choose among those words the one who, he thinks, give the highest result - or, alternatively, which gives the two best, or still the three better, that is to say that in the latter case the process amounts to classifying the words by decreasing result. We can also ask the user to select or classify three, four or five words among ten or twenty that are proposed to him.

Answers can be weighted by a markup even more important that the selection and / or ranking are relevant, and vice versa.

In another aspect, the invention can be applied to the analysis either a succession of words considered in isolation, but of a set of words considered collectively. It is indeed difficult to evaluate intuitively, let alone to quantify, global relevance of a group of words, and to distinguish a group of words close to sense, of a group of words far from meaning (the "meaning" being defined by the word departure) It is then necessary to estimate, with respect to a given starting word, the relevance relative overall, of two (or more) lists of n words (for example n = 10 words), that is, say of determine the one that is the richest semantically.

-13-To do this, a "match index" is calculated for each list, which is a clue representative of its relevance to the original word. This index of comparison will then make it possible to compare various lists on the other hand, eliminate or retain a given list by comparing the index to a threshold given, which threshold may itself be parameterized according to the degree of requirement required to the discrimination.

We will give an example of the implementation of this technique, with reference to Tables 5 and 6 below. In this example, the starting word is electronic.
From this starting word the search engine gives a result of 9,200,000 occurrences.

The first step is to establish two lists of additional words, for example two lists of ten words (see Table 5 for the first list and Table 6 for the second listing).

In the example in Table 5, the first list includes the words: memory, relay, register, ... multiplexer, Zener. Similarly, the second list in Table 6 includes words comparator, gate, resistor ... transformer, capacitor.

The next step is to rank the different additional words of each list, by decreasing result. Both lists are presented in this way, the result being expressed as a relative percentage (18.67%, 13.02%, ...) compared to the result of the word departure (100%).

-14-Start word Additional word% relative Results electronic 100.0000 9 210 000 electronic memory 18,6754 1,720,000 relay electronics 13,0293 1,200,000 electronic register 12,5950 1,160,000 electronic meter 11,8350 1,090,000 electronic decoder 7,2096 664,000 electronics integrator 4,1151 379,000 electronic shielding 2,6602 245,000 electronic buffer 1.6830 155 000 multiplexer electronics 0,9522 87,700 electronic zener 0.5364 49 400 Average of 10 words: 7,329 Related index (Max / Avg) 2.55 If the threshold is 2.5, then the index of relationship is too high relevant Table 5. First list of ten words for the example

-15-Start word Additional word% relative Results electronic 100.0000 9 210 000 electronic comparator 19,0011 1,750,000 electronic door 17,5896 1,620,000 electronic resistance 13,8979 1,280,000 electronic amp 13,4636 1,240,000 electronic self 11.4007 1 050 000 electronic diode 7,4593 687,000 electronic decoder 7,2096 664,000 electronics radiator 5,5809 514,000 electronic transformer 5,1574 475,000 electronic capacitor 4,8317 445,000 Average of 10 words: 10,559 Index of Relation (Max / Avg) 1.80 If the threshold is 2.5, index of relatedness below the threshold so list relevant Table 6. Second list of ten words for the example

-16-The next step is to calculate the average of the ten words (or ten best words, if more words were tested), or 7.329% for the first list and 10,559% for the second list. This average could be retained as index related to the relevance of the first list (average:
10,559) by report to the second list (average: 7.329).

Experience shows, however, that in practice this criterion does not optimize always Relative relevance assessment for this list.

The invention proposes to further improve discrimination by applying a weighting.
This weighting may in particular be a function of the highest result obtained for different words from the list, that is to say according to the result of the "better" additional word.

The index of relationship can be defined for this purpose as the quotient result obtained with the best additional word of the list (MAc) by the average of all the words of this same list (MOY).

In the example of the two lists in Tables 5 and 6, for the first list index thus calculated is: 18.6754 = 7.329 = 2.55, while for the second list it is worth: 19,0011 = 10,559 = 1,80.

The index thus calculated is a number greater than unity, and the relevance of the list will be even higher than the value of this index will be low (close to unity), and vice versa. Note that if the index was calculated in the opposite way, that is, say MOY / MAX, its value would be between 0 and 1, high relevance being reflected by a value high, close to the unit.

In all cases, a value close to unity indicates a strong relevance, revealing words all with high scores, close to the maximum score.

Advantageously, the index of comparison is compared with a given threshold, fixed according to selectivity sought: a threshold close to unity will correspond to a requirement high on the criterion of relevance, while a higher threshold to consider as relevant a larger number of lists.

-17-With the example of Tables 5 and 6, we see that if we set the threshold at 2.5, the first list will be considered as insufficiently relevant (its index relatedness is 2.55), while the second list will be much more relevant (index relatedness: 1.80).

The threshold is very advantageously parameterizable by the user, for example via a graphical interface including arrows or potentiometer slider linear.

Whether or not the relevance test is verified, that is, the position of the index of each list compared to the threshold, can be displayed on this same interface immediately noticeable in the form of an indicator Boolean, by example, a change of color or the display of a "RELEVANT" message in a cell.

If we take the example of Tables 5 and 6, setting the threshold at 2.5 causes the display the "RELEVANT" message for the second list, and for this list only, because the index related to the first list (2.55) is too high compared to threshold (2.5) of so that this first list is considered irrelevant.

If the user modifies the threshold, for example by increasing it from 2.5 to 2.6, then he will see display the "RELEVANT" message for the first list as well.
Conversely if lower the threshold, for example from 2.5 to 1.7, the message "RELEVANT" will disappear for the two lists.

Such an interactive display offers the user the opportunity to react directly thanks to the immediate perception of the result achieved, variable depending on the threshold setting, allowing a particularly efficient implementation of the method of the invention.

The weighting of the index of relationship may also take into account, in a more elaborate variant, of the difference between the result of the starting word and the average corresponding of the different words of the list.

Such weighting makes it possible, in particular, to take into account the fact that, for a word from departure giving a low initial result (in absolute terms), it will be easy to find words additions giving close results (in relative value) of this result initial,

-18-leading to a relatively favorable index, due to the low standard deviation of words additional. On the other hand, for a starting word giving an initial result high he will difficult to find additional words giving results close to this result thus leading to a relatively poor index. Taking into account of the gap between, on the one hand, the result of the starting word and, on the other hand, the average results additional words will reflect this increased difficulty and will give a value weighted index.

It is also possible to take into account the number of words in the list, considered as a variable and no longer as a fixed parameter at the beginning. In that case, index related will be weighted so as to appear even better than the number of words in the list is high: it is indeed easier to find a short relevant list than a long list, because the probability then increases of meeting little terms relevant, likely to reduce the average. This increased difficulty is taken into count if the index is weighted by the number of additional words in the list.

An index of this type is for example calculated from the formula:
INDEX = [(MAX / MOY)] * N, in which MAx is the result obtained with the best additional word, MOY being the average of the results of all the words in the list, Where N is the number of words in the list.

In general, there are many ways to operate the various weights that we have just exposed.

-19-By way of non-limiting example, it is possible to calculate an index of from the following generalized formula:

INDEX = [1 / (MDD- MAX)] * [1 / (MDD- MOY)] * N, where MDD is the result of the word of departure, MAX being the result obtained with the best additional word, MOY being the average of the results of all the words in the list, Where N is the number of words in the list.

In this index:

the factor [1 / (MDD - MAx)] is representative of the quality of the various words additional found ; it reflects the difficulty of establishing a relevant list when the "best"
additional word has a high result;

the factor [1 / (MDD - MOY)] is representative of the consistency of the results various additional words; it reflects the low dispersion of these words around the average ;

the factor N, corresponding to the length of the list, reflects the character more or less prolix process of generating additional words.

This weighted index will allow to classify several lists of words, from the most significant to the least significant, that is, from the richest to the less rich on the map semantics.

Numerous variants of calculation of this index can be of course considered, for example starting from the median instead of the average, in introducing from nonlinearities by elevations to full or fractional power from some terms, by the application of thresholds or capping, etc., so as to optimize the relevance of the index according to the results of the experiments practice.

-20-

Claims (18)

1. A method for analyzing and processing requests applied to an engine of research, this method being implemented by a computer system including said search engine, means to produce vocabulary words, means to submit to the search engine queries each containing a word, or a plurality of associated words grouped in one sentence, and means for receiving in response to each of these queries a numerical result representative of occurrences the word, or phrase, in a set of web pages indexed by the engine of search, this method being characterized in that it comprises the sequence steps successive of:

a) determination of a starting word and submission of a request containing that word of departure ;

b) receiving an initial result corresponding to this request containing the word of departure;
(c) generation of an additional word and submission of a request containing a phrasing including the word of departure and the additional word generated, distinguished by a separator;

(d) receipt of the result, lower than the initial result, corresponding to the request made to step c);

e) reiteration N-1 times of steps c) and d) with each time a word additional, thus giving N respective results corresponding to the additional N words successively generated; and f) ranking of the N results thus obtained. 2. The method of claim 1, further comprising, after step f), a step of:
g) assignment of a relationship index according to a rule according to said N
results, this index of relatedness being a global index attributed to the family of N
words generated, this index of relatedness being also an index relative, attributed in consideration of the word of departure. 3. The method of claim 2, wherein said index of relatedness is a clue Mean function (MOY) of the results of the additional words. 4. The method of claim 2, wherein said index of relatedness is a clue weighted by the value (MAX) of the result corresponding to the additional word giving the highest result. 5. The method of claim 2, wherein said index of relatedness is a clue weighted by the value (MDD) of the result corresponding to the original word. The method of claim 2, wherein said index of relatedness is a clue weighted by the number N of additional words. 7. The method of claim 3, wherein said index of relatedness is a clue weighted by the value (MAX) of the result corresponding to the additional word giving the highest result and in which said index of relationship is a index of the quotient (MAX / AVERAGE) of the result (MAX) corresponding to the additional word giving the result the highest by the mean (MOY) of the results of the additional words. The method of claim 3, wherein said index of relatedness is a clue weighted by the value (MDD) of the result corresponding to the starting word and which said index of relatedness is an index depending on the difference (MDD -MOY) between value (MDD) of the result corresponding to the starting word and the mean (MOY) of the additional word results. 9. The method of claim 4, wherein said index of relatedness is a clue weighted by the value (MDD) of the result corresponding to the starting word and which said index of relatedness is an index depending on the difference (MDD -MOY) between value (MDD) of the result corresponding to the start word and the value (MAX) of the result corresponding to the additional word giving the highest result. 10. The process of claim 3, wherein said relatedness index is a value-weighted index (MAX) result corresponding to the additional word giving the highest result;

wherein said relatedness index is a value-weighted index (MDD) result corresponding to the word of departure;

wherein said index of match is a number-weighted index of words additional; and wherein said index of relatedness is a function index of the expression:
[1 / (MDD - MAX)] * [1 / (MDD - MOY)] * N, where MDD is the result of the starting word, MAX being the result obtained with the best word additional, MOY being the average of the results of all the words in the list, N being the number of words in the list. The method of claim 2, further comprising, after step g), a step :

h) comparison of the index of relationship to a given threshold, and production a Boolean indicator whose value reflects the crossing or not of this threshold. The method of claim 1, wherein the computer system includes in outraged :

at least one terminal (10, 20) capable of presenting data to a user and allow this user to provide data to the system, this terminal comprising said means for submitting queries to the search engine and for to receive in response a numerical result, and a central site (40) coupled to the terminal or to each of the terminals, and to engine of research. The method of claim 12, wherein step a) of determining of the word departure is a step of selecting a word by a user in a set of words predefined. The method of claim 12, wherein step a) of determining of the word departure is a step of selecting a word by the central site among a set of words predefined. 15. The method of claim 12, wherein the system is implemented works by a plurality of users and said additional words generated in step c) are words provided by users to the system, each user providing the less a word additional amount. 16. The method of claim 12, wherein, the starting word and the words additional additions being generated by the system, the method further comprises, after step f), the following steps:

i) by the central site, presentation to the user of the N additional words, in one order any;

j) by the user, establishing an estimated ranking of M words selected among the N
additional words presented, with M <= N, and communication of that estimated ranking at the central site; and (k) attribution of a relevance score, as a function of the proximity between the ranking estimated release by the user, and the actual ranking determined at step f). 17. The method of claim 12, wherein said terminal is a phone cell (20), and said central site (40) is a remote site connected by means of telecommunication (30, 52, 54) to a plurality of said terminals. 18. The method of claim 12, wherein said terminal is a computer (10), and said central site (40) is a remote site connected by means of telecommunication (14, 16, 50) to a plurality of said terminals.