Method and apparatus for keyword mass generation

Abstract

A method and apparatus in accordance with the invention which, for any given keyword, generates a numeric value that defines keyword relevance based on the number and importance of a keyword's forward link and back link keyword neighbors.

Description

FEDERALLY SPONSORED RESEARCH

Not applicable

SEQUENCE LISTING OF PROGRAM

Not applicable

BACKGROUND OF THE INVENTION

This invention pertains to technology used for data search, particularly data search over the Internet.

In many practical applications such as documents storing, searching, comparative analysis, and reconstruction, it is extremely important to have a solution allowing a user to compare and rank different keywords associated with documents.

Unfortunately such solution does not exist today, and there are many reasons for this.

First of all, the number of keywords used in any language is incredibly large. For example, there are over 200,000 general purpose words and over 500,000 special words and abbreviations in the English language alone. The number of keywords combing two, three, or four words (called terms) in the English language is respectively 5, 10, and 15 orders of magnitude larger.

Second, there are many parameters characterizing keywords from different points of view. Some of them contradict with others. For example, a search engine operator can characterize keywords by a number of available matches, number of assigned advertisements, or keyword CTR (click through rate). An Internet user can characterize keyword by language, complexity, length, or popularity. An advertiser can characterize keywords by biding prices, generated traffic, CPM, or conversion rate.

Third, there is a lack of any theoretical models that can effectively aggregate multiple viewpoints together in the unified working system to practically address the problem of comparing and ranking keywords and terms.

The proposed invention defines a method and apparatus to compute keyword masses based on the invented keyword mass computation technology.

SUMMARY

The main idea of the invention is to create a universal keyword relevance measure number. We will refer to the keyword relevance number as a K-mass. The invention proposes to compute K-mass by aggregating two intrinsic keyword parameters: forward link and back link (also known as “backlink”) strengths. The forward link strength (“SF”) of the keyword K is a numerical coefficient describing how many other keywords are defined by keyword K as its neighbors (keywords close in meaning and/or co-occurance with K), how similar they are to K, and the aggregation of these two factors for all of K's neighbors (also known as “strength”). The back link strength (“SB”) of a keyword K is a numerical coefficient describing how many other keywords define K as their neighbor, K's strength, and K's similarity to them.

In other words, K-mass of keyword K is an aggregated strength of keywords that K can “see” and keywords that can “see” K. The measure of similarity between a keyword K and its neighbor keyword L is also called the “proximity score” of L relative to K. In one embodiment of this invention proximity score of L relative to K is the inverse of the rank of L in an ordered list of K's neighbor keywords. We will also refer to a neighbor keyword as a “neighbor” below.

In one embodiment of the invention forward and back links aggregation is a function A(N, SF, SB) of the number of words (terms) in the keyword (1, 2, 3, . . . ), and the strengths of SF and SB.

In one embodiment of the invention SF and SB are each computed as a number of appropriate neighbor keywords—SF is computed as a number of neighbor keywords of K, and SB is computed as a number of keywords that have K as a neighbor.

In one embodiment of the invention the back links aggregation function is defined as follows:

A(N,SF,SB)=f(α×g_F(N)×h_F(SF), (1−α)×g_B(N)×h_B(SB)),

where

• • SF=Σr_i^f×sf_i;
  • SB=Σr_i^b×sb_i;
  • α—is a forward link strength weight coefficient;
  • r_i^f—is proximity between keyword K and its forward link neighbor number “i”;
  • r_i^b—is proximity between keyword K and its back link neighbor number “i”;
  • sf_i—is strength of the forward link neighbor number “i”;
  • sb_i—is strength of the back link neighbor number “i”;
  • h_F(SF)—is a monotonically increasing function such as h_F(x)≧1, h_F(1)=1;
  • h_B(SB)—is a monotonically increasing function such as h_B(x)≧1, h_B(1)=1;
  • g_F(N), g_B(N)—non-decreasing functions of number of keyword terms such as g_F(1)=1, g_B(1)=1;
  • f(.)=general average function. A function F(x₁ . . . x_n) is called general average when min(x₁ . . . x_n)≦F(x₁ . . . x_n)≦max(x₁ . . . x_n).

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1—shows a preferred embodiment system block diagram.

DETAILED DESCRIPTION

This invention is related to FIG. 1 which describes the preferred embodiment of the invention. In FIG. 1, keyword 100 is entered into a system to generate its K-mass. Block 110 generates or finds in a database a set of forward link keywords and their forward link proximity scores 130. Block 120 generates or finds in a database a set of back link keywords and their back link proximity scores 140. Block 150 generates keyword's K forward link strength 170. Block 160 generates keyword's K back link strength 180. Block 190 aggregates forward link strength 170 and back link strength 180 and produces a final keyword's K-mass.

For example, let's assume that keyword K has three forward neighbors with relevance scores and strengths (1, 300), (0.90, 450), (0.7, 400). Let's assume that it has four back link neighbors with relevance and strengths (1, 440), (0.95, 470), (0.60, 350), (0.50, 600). Defining weigh coefficient as 0.5 and g_F(2)=4, g_B(2)=4, h_F(x)=h_B(x)=ln(1+x), and f(.) as an arithmetic mean we will compute K-mass as

$\begin{array}{c}A=4(0.5\ln \left(1+300+0.9*450+0.7*400\right)\\ +0.5\ln \left(1+440+0.95*470+0.6*350+0.5*600\right))\\ =40.79\end{array}$

Additional Embodiments

In one embodiment of the invention K-mass is generated using additional information about a keyword's popularity (average number of searches per day) and a keywords click through rate.

In one embodiment of the invention strength of a forward link neighbor is defined as the number of its neighbors.

In one embodiment of the invention the strength of the back link neighbor is defined as a number of its neighbors.

In one embodiment of the invention h_F(.) and h_B(.) are defined as follows:

h_F(x)=1+a_F×ln(x), h_B(x)=1+a_B×ln(x), where a_F and a_B are positive numbers

Although the above description contains much specificity, the embodiments described above should not be construed as limiting the scope of the invention but rather as merely illustrations of some presently preferred embodiments of this invention.

Claims

1. A method of computing a keyword's relevance as a proximity score, comprising: (i) computing a strength of a keyword based on a keyword strength criteria,(ii) computing a list of forward link neighbor keywords for a keyword and proximity scores of said neighbor keywords relative to said keyword,(iii) computing a list of back link neighbor keywords for a keyword and proximity scores of said keyword relative to said neighbor keywords,(iv) computing a keyword relevance measure for a keyword as an aggregation function of at least one of the following: a strength of said keyword, a strength of a forward link neighbor keyword to said keyword, a strength of a back link neighbor keyword to said keyword, a proximity score of a forward link neighbor keyword relative to said keyword, a proximity score of a keyword relative to said keyword's back link neighbor keyword, a measure of word quantity in said keyword, a measure of word quantity in a forward link neighbor keyword for said keyword, and a measure of word quantity in a back link neighbor keyword for said keyword.

2. The method of claim 1 wherein said proximity score is a function of a neighbor keyword rank in an ordered list of neighbor keywords.

3. The method of claim 1 wherein a keyword's mass is computed as follows: (i) a forward link strength of said keyword is computed as a function of at least one of the following: a strength of said keyword, a strength of a forward link neighbor of said keyword, a measure of word quantity in said keyword, and a measure of word quantity in a forward link neighbor of said keyword,(ii) A back link strength of a keyword is computed as a function of at least one of the following: a strength of said keyword, a strength of a back link neighbor of said keyword, a measure of word quantity in said keyword, and a measure of word quantity of a forward link neighbor of said keyword;(iii) A keyword's mass is computed as a function of at least one of the following:a strength of said keyword, a strength of a forward link keyword of said keyword, and a strength of a back link keyword of said keyword.

4. The method of claim 3 wherein said forward link strength is generated using a pre-defined list of forward link neighbor keywords and proximity scores.

5. The method of claim 3 wherein said back link strength is generated using a pre-defined set of back link neighbor keywords and proximity scores.

6. The method of claim 3 wherein said forward link strength is generated using forward link strengths of forward link neighbor keywords of said keyword.

7. The method of claim 3 wherein said back link keyword strength is generated using back link strengths of back link neighbor keywords of said keyword.

8. The method of claim 3 wherein said forward link strength is generated using a measure of keyword quantity on said keyword's forward link neighbor keywords.

9. The method of claim 3 wherein said back link keyword strength is generated using a measure of keyword quantity on said keyword's back link neighbor keywords.

10. The method of claim 3 wherein said forward link and back link strengths of said keyword are generated with human interactions

11. An apparatus comprising: keyword mass generation pipeline, and a forward link neighbor keyword collection and proximity score generation means for said keyword mass generation pipeline, and a back link neighbor keyword collection and proximity score generation means for said keyword mass generation pipeline, and a forward link strength generation means for said keyword mass generation pipeline, and a back link strength generation means for said keyword mass generation pipeline, and a strength computation and aggregation means for said keyword mass generation pipeline.

12. The apparatus of claim 11 wherein said strength generation means includes one or more of the following: a weight coefficients aggregation means, and a keyword rank aggregation means.