Product search method and system

• US 9,563,665 B2
• Filed: 05/15/2013
• Issued: 02/07/2017
• Est. Priority Date: 05/22/2012
• Status: Active Grant

• Current Assignee: Alibaba Group Holding Ltd.
• Original Assignee: Alibaba Group Holding Ltd.
• Inventors: Chen, Chao, Han, Xiaomei, Song, Chao, Wei, Hui
• Primary Examiner(s): BOWEN, RICHARD L
• Application Number:
  US13/894,827
• Publication Number:
  US 20130318101A1
• Time in Patent Office: 1,364 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/3323 using document space presentation or visualization, e.g. category, hierarchy or range presentation and selection
• CPC Class Codes: G06F 16/3326 using relevance feedback from the user, e.g. relevance feedback on documents, documents sets, document terms or passages
• CPC Class Codes: G06F 16/3334 Selection or weighting of terms from queries, including natural language queries
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/38 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06Q 30/0627 using item specifications

Abstract

Embodiments of the present application relate to a search method, a system for searching, and a computer program product for searching. A search method is provided. The method includes receiving a query word string, retrieving a plurality of various product information entries related to the query word string, extracting, from a memory, a first core product word from the query word string, extracting, from the memory, second core product words from the various product information entries, for each second core product word, checking a list of candidate product words relating to a key product word for a presence of the each second core product word, in the event that the each second core product word is present, reducing a weighting of the product information entry corresponding to the each second core product word, and sorting the product information entries according to the adjusted weightings of the product information entries.

Claims

1. A search method, comprising:

receiving a query word string;

retrieving a plurality of product information entries related to the query word string;

extracting, from a memory, a first core product word from the query word string to obtain a first key product word, the first core product word being a smallest semantic unit;

extracting, from the memory, a plurality of second core product words from the product information entries, the plurality of second core product words being smallest semantic units, wherein the extracting of the plurality of second core product words comprises;

extracting last product words from the plurality of product information entries, the plurality of second core product words including the last product words, the plurality of product information entries corresponding to titles of products, descriptions of products, or a combination thereof;

for a first second core product word, checking a list of candidate product words relating to the first key product word for a presence of the first second core product word;

in the event that the first second core product word is present in the list of candidate product words, reducing a weighting of the product information entry corresponding to the first second core product word, wherein the list of candidate product words contains a first candidate product word, the first candidate product word relating to the first second core product word, wherein the first candidate product word is combined with the first key product word to obtain a synthesized product word, and wherein the first key product word and the synthesized product word do not belong to the same category; and

sorting and outputting the product information entries according to the reduced weightings of the product information entries.

2. The method as described in claim 1, further comprising:

establishing the list of candidate product words comprises;

for at least one product information entry contained in a database;

performing a coarse granularity segmentation by the largest semantic units; and

extracting a third core product word contained in segmented results;

determining whether the third core product word has been extracted from the segmented results;

in the event that the third core product word has been extracted from the segmented results, performing a fine granularity segmentation by the smallest semantic units;

determining whether at least two of the words obtained are product words;

in the event that at least two of the words obtained are product words;

using the first product word as a key product word; and

using the last product word as a candidate product word of the key product word;

computing correlations of at least one key product word and at least one candidate product word;

determining whether the correlation of the at least one key product word and the at least one candidate product word meets a threshold value;

selecting a candidate product word having a correlation that meets the threshold value; and

for the same key product word, generating the list of candidate product words based on the selected candidate product word.

3. The method as described in claim 2, wherein the computing of the correlations of the at least one key product word and the at least one candidate product word, the determining whether the correlation of the at least one key product word and the at least one candidate product word meets the threshold value, and the selecting of the candidate product word having the correlation that meets the threshold value comprises:

for the at least one key product word and the at least one candidate product word;

vectorizing the at least one key product word based on a click through rate for a category of the at least one key product word to obtain a vector; and

vectorizing the at least one candidate product word based on a click through rate for a category of the at least one candidate product word to obtain a vector;

computing angle values between the vectors corresponding to the key product words and the vectors corresponding to the candidate product words;

computing correlations between the angle values;

determining whether at least one candidate product word having a correlation meets a threshold value based on an angle value; and

selecting a candidate product word having the correlation that meets the threshold value based on the angle value.

4. The method as described in claim 3, wherein for the at least one key product word and the at least one candidate product word, the vectorizing of the at least one key product word using the click through rate for the category of the key product word to obtain the vector, and the vectorizing of the at least one candidate product word using the click through rate for the category of the at least one candidate product word to obtain the vector comprises:

for the at least one key product word and the at least one candidate product word, from a search log, separately analyzing and tabulating the click through rates of respective relevant categories to obtain a list of weightings for a relevant category;

converting various values in the list of weightings of the at least one key product word into the vector; and

converting various values in the list of weightings of the at least one candidate product words into the vector.

5. The method as described in claim 4, wherein the computing of the angle values between the vectors corresponding to the at least one key product word and the vectors corresponding to the at least one candidate product word, and the selecting of the candidate product word having the correlation that meets the threshold value based on the angle value comprises:

computing a cosine angle value of the vectors;

determining whether the cosine angle value is greater than the threshold value; and

in the event that the cosine angle value is greater than the threshold value, deleting the candidate product word.

6. The method as described in claim 2, wherein the using of the last product word as the candidate product word of the key product word comprises:

setting the last product word as a second core product word of a relevant product information entry; and

storing the corresponding product information entry.

7. The method as described in claim 6, wherein the extracting of the second core product words corresponding to the product information entries comprises:

extracting the second core product words of the product information entries from the memory.

8. The method as described in claim 2, wherein:

the computing correlations of the at least one key product word and the at least one candidate product word comprises;

vectorizing the at least one key product word based on a category click through rate, an attribute click through rate, and a product word click through rate of the each key product word;

vectorizing the at least one candidate product word based on the category click through rate, the attribute click through rate, and the product word click through rate of the at least one candidate product word; and

computing an angle value between the vector corresponding to the at least one key product word and the vector corresponding to the at least one candidate product word; and

the selecting of the candidate product word having the correlation that meets the threshold value comprises;

determining whether the at least one candidate product word has the correlation meeting the threshold value based on the obtained angle value; and

selecting the candidate product word having the correlation meeting the threshold value.

9. The method as described in claim 1, wherein the extracting of the second core product words comprises:

parsing the query word string or product information entry text to obtain parsed words;

labeling word characteristics of the parsed words;

checking a list of contextually irrelevant words having word characteristics of operator words to identify coordinating conjunctions contained in the list of contextually irrelevant words;

segmenting the query word string or product information entry text based on the coordinating conjunctions; and

for each segment of the query word string;

determining whether the each segment contains at least two parsed words which have a word characteristic of a product word; and

in the event that the each segment contains the at least two parsed words which have the word characteristic of the product word, setting the word characteristic of the last of these parsed words as a core product word.

10. A search method, comprising:

receiving a query word string;

retrieving a plurality of product information entries related to the query word string;

extracting, from a memory, a first core product word from the query word string to obtain a first key product word, the first core product word being a smallest semantic unit;

extracting, from the memory, a plurality of second core product words from the product information entries, the plurality of second core product words being smallest semantic units;

for each second core product word, checking a list of candidate product words relating to the first key product word for a presence of a first second core product word;

in the event that the each second core product word is present in the list of candidate product words, reducing a weighting of the product information entry corresponding to the first second core product word, wherein the list of candidate product words contains a first candidate product word, the first candidate product word relating to the first second core product word, wherein the first candidate product word is combined with the first key product word to obtain a synthesized product word, and wherein the first key product word and synthesized product word do not belong to the same category;

sorting and outputting the product information entries according to the reduced weightings of the product information entries;

establishing the list of candidate product words, comprising;

for each product information entry of product information entries contained in a database;

performing a coarse granularity segmentation by the largest semantic units; and

extracting a third core product word contained in segmented results;

determining whether the third core product word has been extracted from the segmented results;

in the event that the third core product word has been extracted from the segmented results, performing a fine granularity segmentation by the smallest semantic units;

determining whether at least two of the words obtained are product words;

in the event that at least two of the words obtained are product words;

designating the first product word of the product words as a key product word; and

designating the last product word of the product words as a candidate product word of the key product word;

computing correlations of each key product word and each candidate product word, comprising;

for the each key product word and the each candidate product word;

separately vectorizing each key product word using three dimensions of a category click through rate, an attribute click through rate, and a product word click through rate of the each key product word; and

separately vectorizing each candidate product word using the three dimensions of the category click through rate, the attribute click through rate, and the product word click through rate of the each candidate product word; and

for each dimension, computing an angle value between the vector corresponding to the each key product word and the vector corresponding to the each candidate product word for the three dimensions;

determining whether the correlation of the each key product word and the each candidate product word meets a threshold value;

selecting a candidate product word having a correlation that meets the threshold value, comprising;

determining whether the each candidate product word having a correlation meeting the threshold value based on the obtained angle values of the three dimensions; and

selecting a candidate product word having the correlation meeting the threshold value;

for the same key product word, generating the list of candidate product words based on the selected candidate product word.

11. The method as described in claim 10, wherein for the each key product word and the each candidate product word, the vectorizing of the each key product word using the three dimensions of the category click through rate, the attribute click through rate, and the product word click through rate of the each key product word, and the vectorizing of the each candidate product word using the three dimensions of the category click through rate, the attribute click through rate, and the product word click through rate of the each candidate product word comprises:

for the each key product word and the each candidate product words;

analyzing and retrieving a search log; and

retrieving the product information entries clicked via the each key product word and the each product information entry clicked via the candidate product word;

tabulating a first category click through rate list corresponding to the key product words, a corresponding first attribute click through list of attributes of the product information entries, and a corresponding first product word click through rate list of the core product words of the product information entries;

tabulating a second category click through rate list corresponding to the candidate product words, a corresponding second attribute click through rate list of attributes of the product information entries, and a second product word click through rate list of the core product words of the product information entries; and

converting the first category click through rate list, the first attribute click through rate list, the first product word click through rate list, the second category click through rate list, the second attribute click through rate list, and the second product word click through rate list into corresponding vectors.

12. The method as described in claim 11, wherein for each dimension, the computing of the angle value between the vector corresponding to the each key product word and the vector corresponding to the each candidate product word, and the selecting of the candidate product word having the correlation that meets the threshold value based on the obtained angle values of the three dimensions comprises:

computing a cosine angle value between the vector corresponding to the first category click through rate list and the vector corresponding to the second category click through rate list to obtain a category click through rate correlation;

computing a cosine angle value between the vector corresponding to the first attribute click through rate list and the vector corresponding to the second attribute click through rate list, to obtain an attribute click through rate correlation;

computing a cosine angle value between the vector corresponding to the first product word click through rate list and the second product word click through rate list to obtain the product word click through rate correlation;

linearly weighting of the category click through rate correlation, the attribute click through rate correlation, and the product word click through rate correlation to obtain an overall similarity of the key product words and the candidate product words;

determining whether the overall similarity is greater than a threshold value; and

in the event that the overall similarity is greater than the threshold value, deleting the candidate product words.

13. A system for searching, comprising:

at least one processor configured to;

receive a query word string;

retrieve a plurality of product information entries related to the query word string;

extract a first core product word from the query word string to obtain a first key product word, the first core product word being a smallest semantic unit;

extract a plurality of second core product words from the product information entries, the second core product words being smallest semantic units, wherein the extracting of the plurality of second core product words comprises to;

extracting last product words from the plurality of product information entries, the plurality of second core product words including the last product words, the plurality of product information entries corresponding to titles of products, descriptions of products, or a combination thereof;

for a first second core product word, check a list of candidate product words relating to the first key product word for a presence of the first second core product word;

in the event that the first second core product word is present in the list of candidate product words, reduce a weighting of the product information entry corresponding to the first second core product word, wherein the list of candidate product words contains a first candidate product word, the first candidate product word relating to the first second core product word, wherein the first candidate product word is combined with the first key product word to obtain a synthesized product word, and wherein the first key product word and the synthesized product word do not belong to the same category; and

sort and output the product information entries according to the reduced weightings of the product information entries; and

a memory coupled to the at least one processor and configured to provide the at least one processor with instructions.

14. The system as described in claim 13, wherein the at least one processor is further configured to:

establish the list of candidate product words comprises;

for at least one product information entry contained in a database;

perform a coarse granularity segmentation by the largest semantic units; and

extract a third core product word contained in segmented results;

determine whether the third core product word has been extracted from the segmented results;

in the event that the third core product word has been extracted from the segmented results, perform a fine granularity segmentation by the smallest semantic units;

determine whether at least two of the words obtained are product words;

in the event that at least two of the words obtained are product words;

use the first product word as a key product word; and

use the last product word as a candidate product word of the key product word;

compute correlations of at least one key product word and at least one candidate product word;

determine whether the correlation of the at least one key product word and the at least one candidate product word meets a threshold value;

select a candidate product word having a correlation that meets the threshold value; and

for the same key product word, generate the list of candidate product words based on the selected candidate product word.

15. A computer program product for searching, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

receiving a query word string;

retrieving a plurality of product information entries related to the query word string;

extracting, from a memory, a first core product word from the query word string to obtain a first key product word, the first core product word being a smallest semantic unit;

extracting, from the memory, a plurality of second core product words from the product information entries, the second core product words being smallest semantic units, wherein the extracting of the plurality of second core product words comprises;

extracting last product words from the plurality of product information entries, the plurality of second core product words including the last product words, the plurality of product information entries corresponding to titles of products, descriptions of products, or a combination thereof;

for a first second core product word, checking a list of candidate product words relating to the first key product for a presence of the first second core product word;

in the event that the first second core product word is present in the list of candidate product words, reducing a weighting of the product information entry corresponding to the first second core product word, wherein the list of candidate product words contains a first candidate product word, the first candidate product word relating to the first second core product word, wherein the first candidate product word is combined with first key product word to obtain a synthesized product word, and wherein the first key product word and the synthesized product word do not belong to the same category; and

sorting and outputting the product information entries according to the reduced weightings of the product information entries.