IP Library › Granted Patent US 12,664,219
Granted Patent B2
US 12,664,219 · App. 18/818,512 · Granted Jun 23, 2026

System and method for topic extraction and opinion mining

Inventors: Neelakantan Sundaresan (Mountain View, CA); Yongzheng Zhang (San Jose, CA); Catherine Baudin (Palo Alto, CA); Dan Shen (Shanghai, CN); Shen Huang (Shanghai, CN)
Assignee: eBay Inc.
G06F16/93G06F16/334G06F16/3344G06F16/36G06F16/5846G06F40/30G06Q10/00G06Q30/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,664,219
App. No.
18/818,512
Granted
Jun 23, 2026
Kind
B2
Abstract

Methods, apparatus, and systems to determine a niche market of items or services, the first phase of which identifies a gap between demand and supply for a set of items. Session logs may be evaluated to compare transactions involving a specific item to those of a larger group of items. The resultant information identifies areas of high demand, but with low availability. The niche market information may be provided as direct merchandising items for sellers. In one example, the method generates niche market item web pages in specific categories. Additional methods, apparatus, and systems are disclosed.

Claims (60)

1 . A method comprising:

accessing a plurality of target documents obtained from one or more sources;

identifying key phrases from the plurality of target documents, the identifying being based on a key phrase extraction process that builds a model based on training documents and uses the model to predict a likelihood of each phrase in a target document of the plurality of target documents;

ranking the key phrases by applying a ranking algorithm;

identifying essential key phrases having rankings that exceed a threshold;

using the essential key phrases as input to Latent Dirichlet Allocation (LDA), generating, using LDA, a list of essential topics, input data for LDA being formed from the essential key phrases selected based on the rankings exceeding the threshold;

associating each of the plurality of target documents with one or more of the essential topics based on essential key phrases found in each of the plurality of target documents; and

outputting the list of essential topics and their association with the plurality of target documents to an analyzer for use in further automated processing.

2 . The method of claim 1 , wherein;

the ranking algorithm is computed as a function of Term Frequency-Inverse Document Frequency (TF-IDF) weights.

3 . The method of claim 1 , further comprising:

assigning an impact score to one or more words of the plurality of target documents, the impact score identifying an impact of the one or more words on a given topic of the essential topics.

4 . The method of claim 3 , further comprising:

identifying one or more opinion trends for at least one essential topic based on the impact score assigned to the one or more words of the plurality of target documents by identifying a number of words for each opinion trend type over time for the at least one essential topic; and

causing display of a user interface on a further client device that comprises one or more graphical indicators of the one or more opinion trends for the at least one essential topic.

5 . The method of claim 3 , further comprising:

causing display of a user interface that illustrates the impact score of the one or more words.

6 . The method of claim 3 , further comprising:

determining that a word of the one or more words is a polarity word based on matching the word to another word in a dictionary of polarity words.

7 . The method of claim 3 , wherein the assigning of the impact score to the one or more words of the plurality of target documents is based on a sum of polarities method.

8 . The method of claim 3 , wherein the assigning of the impact score to the one or more words of the plurality of target documents is based on a syntactic distance between the one or more words and one of the essential key phrases in a syntactic tree.

9 . The method of claim 1 , further comprising:

grouping target documents of the plurality of target documents based on occurrence and use of the essential key phrases found in each of the plurality of target documents.

10 . The method of claim 1 , further comprising:

identifying sub-topics associated with the essential topics; and

extending the list of essential topics to include the associated sub-topics.

11 . A system comprising:

one or more hardware processors; and

a memory storing instructions that, when executed by the one or more hardware processors, cause the one or more hardware processors to perform operations comprising:

accessing a plurality of target documents obtained from one or more sources;

identifying key phrases from the plurality of target documents, the identifying being based on a key phrase extraction process that builds a model based on training documents and uses the model to predict a likelihood of each phrase in a target document of the plurality of target documents;

ranking the key phrases by applying a ranking algorithm;

identifying essential key phrases having rankings that exceed a threshold;

using the essential key phrases as input to Latent Dirichlet Allocation (LDA), generating, using LDA, a list of essential topics, input data for LDA being formed from the essential key phrases selected based on the rankings exceeding the threshold;

associating each of the plurality of target documents with one or more of the essential topics based on essential key phrases found in each of the plurality of target documents; and

outputting the list of essential topics and their association with the plurality of target documents to an analyzer for use in further automated processing.

12 . The system of claim 11 , wherein the operations further comprise:

assigning an impact score to one or more words of the plurality of target documents, the impact score identifying an impact of the one or more words on a given topic of the essential topics.

13 . The system of claim 12 , wherein the operations further comprise:

identifying one or more opinion trends for at least one essential topic based on the impact score assigned to the one or more words of the plurality of target documents by identifying a number of words for each opinion trend type over time for the at least one essential topic; and

causing display of a user interface that comprises one or more graphical indicators of the one or more opinion trends for the at least one essential topic.

14 . The system of claim 12 , wherein the operations further comprise:

causing display of a user interface that illustrates the impact score of the one or more words.

15 . The system of claim 12 , wherein the operations further comprise:

determining that a word of the one or more words is a polarity word based on matching the word to another word in a dictionary of polarity words.

16 . The system of claim 12 , wherein the assigning of the impact score to the one or more words of the plurality of target documents is based on a sum of polarities method.

17 . The system of claim 12 , wherein the assigning of the impact score to the one or more words of the plurality of target documents is based on a syntactic distance between the one or more words and one of the essential key phrases in a syntactic tree.

18 . The system of claim 11 , wherein the operations further comprise:

grouping target documents of the plurality of target documents based on occurrence and use of the essential key phrases found in each of the plurality of target documents.

19 . The system of claim 11 , wherein the operations further comprise:

identifying sub-topics associated with the essential topics; and

extending the list of essential topics to include the associated sub-topics.

20 . A non-transitory computer-readable medium storing instructions that, when executed by one or more hardware processors, cause the one or more hardware processors to perform operations comprising:

accessing a plurality of target documents obtained from one or more sources;

identifying key phrases from the plurality of target documents, the identifying being based on a key phrase extraction process that builds a model based on training documents and uses the model to predict a likelihood of each phrase in a target document of the plurality of target documents;

ranking the key phrases by applying a ranking algorithm;

identifying essential key phrases having rankings that exceed a threshold;

using the essential key phrases as input to Latent Dirichlet Allocation (LDA), generating, using LDA, a list of essential topics, input data for LDA being formed from the essential key phrases selected based on the rankings exceeding the threshold;

associating each of the plurality of target documents with one or more of the essential topics based on essential key phrases found in each of the plurality of target documents; and

outputting the list of essential topics and their association with the plurality of target documents to an analyzer for use in further automated processing.