# US 9,519,936 B2 · Method and apparatus for analyzing and applying data related to customer interactions with social media

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 9,519,936 B2
- Document ID: US9519936B2
- Status: Granted
- Kind: B2
- Application number: 13/461,631
- Filed: 2012-05-01
- Published: 2012-09-13
- Granted: 2016-12-13
- Earliest claimed: 2011-01-19
- Assignees: 24/7 Customer, Inc.
- Applicants: Ravi Vijayaraghavan; Ravi Garikipati; Andrew Chang; Pallipuram V. Kannan
- Inventors: Ravi Vijayaraghavan; Ravi Garikipati; Andrew Chang; Pallipuram V. Kannan
- Web page: https://patentgrep.com/patent/US9519936B2

## CPC

- G06Q 30/02: Marketing; Price estimation or determination; Fundraising
- G06F 16/2465: {Query processing support for facilitating data mining operations in structured databases}
- G06Q 10/0637: Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
- G06Q 10/44: Identification of trends within social networks, e.g. identification of trending topics
- G06Q 10/46: Determination of level of influence of users within social networking services
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Embodiments of the invention provide techniques that quantize community interactions with social media to understand and influence consumer experiences.

## Claims

### Claim 1

An apparatus for quantizing community interactions with social media to understand and influence consumer experiences, the apparatus comprising:
a processor configured for analyzing posts on social media for use in customer service to improve brand and customer experience;
said processor configured for determining one or more leading indicators of issues that appear in a consumer contact channel by
mining said posts,
extracting topics from said posts,
selecting those topics that are occurring for a first time or whose frequency of occurring is changing over time,
calculating a term frequency inverse document frequency (TFIDF) score for each feature in each selected topic,
assigning a weight to each TFIDF score,
calculating a score for each of said posts by summing the weighted TFIDF scores of all features within each post, and
selecting top posts from each selected topic;
said processor configured for predicting a trend in future consumer contact channel activity based on said leading indicators and the analysis of said posts;
said processor configured for identifying one or more key influencers from social media; and
said processor configured for, once said key influencers are identified, contacting said key influencers to influence said key influencers.

### Claim 2 (depends on 1)

The apparatus of claim 1, said processor configured for:
using social media for any of identifying a spike in activity, identifying questions around the spike, or, in a call center application, determining plan staffing and planning an issue response in advance.

### Claim 3 (depends on 1)

The apparatus of claim 1, said processor configured for:
applying a model to quantify the influence of said key influencers.

### Claim 4 (depends on 1)

The apparatus of claim 1, said processor configured for:
determining an influencer trust score for each of said key influencers, where Trust Score=w1(agreement)+w2(Opposition)+w3(retweets)+w4(References), and wherein the weights are optimized based on a manual ranking of each attribute.

### Claim 5 (depends on 4)

The apparatus of claim 4, said processor configured for:
tracking a subset of said key influencers in a community on a social medium, wherein the subset is composed of those influencers having the highest influencer trust scores;
measuring customer sentiments; and
if a negative trend in customer sentiment is identified, contacting customers with a resolution message to create a positive trend.

### Claim 6 (depends on 5)

The apparatus of claim 5, wherein the subset of said key influencers is determined by aggregating the influencer trust score for each influencer;
wherein each of said influencer trust scores is a numerical sum of instances of a particular attribute; and
wherein said influencer trust scores are normalized to an average number of posts from posters in that population.

### Claim 7 (depends on 5)

The apparatus of claim 5, said processor configured for:
providing a model for scoring said key influencers, in which every post is weighted based upon the influence of a person making the post to yield a net impact score;
wherein factors that make up influence include any of engagement, trust, or a poster's network.

### Claim 8 (depends on 5)

The apparatus of claim 5, said processor configured for:
identifying influence that an influencer exerts over others; and
providing a path to remediation of an influencer's issues that creates a net media positive trend in social media.

### Claim 9 (depends on 5)

The apparatus of claim 5, said processor configured for:
monitoring a particular key influencer on a company's social media; and
mining said particular key influencer's posts to provide a service anywhere function that provides service in a channel of said particular key influencer's choice.

### Claim 10 (depends on 1)

The apparatus of claim 1, said processor configured for:
using knowledge of said leading indicators and said key influencers to respond preemptively, either through a particular key influencer and/or through preparing for issues arising later in other customer care channels;
wherein the key issues are identified using a topic model and/or an issue categorization model.

### Claim 11 (depends on 1)

The apparatus of claim 1, said processor configured for:
preparing one or more channels to serve a customer better and/or to provide a better predictive experience to the customer for both agent-based customer service and self-service.

### Claim 12 (depends on 1)

The apparatus of claim 1, said processor configured for:
providing service to a customer in a customer's channel of choice when and where the customer needs it by text mining social media.

### Claim 13 (depends on 1)

The apparatus of claim 1, said processor configured for:
providing any of a URL, a proactive chat window, or a guided self-service widget to a user to address a problem and/or an issue that the user is facing and that the user posted about in a channel where the user expressed the issue.

### Claim 14 (depends on 1)

The apparatus of claim 1, said processor configured for:
providing a sentiment analyser that identifies sentiment polarity from social media text and/or chats and that provides a sentiment strength score for a given input text.

### Claim 15 (depends on 14)

The apparatus of claim 14, said sentiment analyzer configured for:
using a supervised approach using labelled data;
wherein a set of posts are manually tagged/labelled as negative and positive sentiment posts;
wherein negative sentiment associated with each word in a post is determined by calculating its normalized likelihood on negative posts; and
wherein positive sentiment associated with each word in a post is determined by calculating its normalized likelihood on positive posts; and
using a Bayesian model to calculate a negative sentiment of a sentence by combining an individual word's sentiment scores;
wherein unigrams and bi-grams are used as features.

### Claim 16 (depends on 14)

The apparatus of claim 14, said sentiment analyzer configured for:
using an unsupervised approach using sentiment dictionaries to provide a sentiment score for each of a list of words which are used to express a sentiment;
wherein in a corpus being analyzed for sentiment, words that are part of a sentiment dictionary are identified and their score is obtained;
wherein scores for all sentiment words in a document or a post are summed up to provide a sentiment score for that particular document or post; and
wherein said sentiment score is normalized by dividing by a total number of words in the document.

### Claim 17 (depends on 1)

The apparatus of claim 1, said processor configured for:
providing a snapshot of consumer sentiment across a broad range of issues that may affect a merchant.

### Claim 18 (depends on 17)

The apparatus of claim 17, said processor configured for:
producing said snapshot by assuming each positive or negative term to be equivalent; and
calculating a score as: (#Positive terms−#Negative terms)/(#Positive terms +#Negative terms).

### Claim 19 (depends on 1)

The apparatus of claim 1, said processor configured for:
using words/features in a long tail of a word/feature statistical distribution to identify unique issues and their trends in chats/social media that are associated with unique features;
wherein frequency of any word is inversely proportional to its rank in a frequency table;
wherein a majority of words and, hence, features have very low frequency; and
wherein unique occurrences in time with low frequency represent interesting phenomena.

### Claim 20 (depends on 1)

The apparatus of claim 1, said processor configured for:
providing an insights engine for identifying discriminatory interaction features;
wherein a database for two juxtaposed features is applied to a feature extraction engine which identifies said juxtaposed features as tri-grams, bi-grams, and uni-grams to which an ensemble of feature importance measures is applied; and
wherein said feature importance measures comprise any of bi-normal separation, information gain, chi-squared, odds ratio, inverse document frequency (IDF), or power.

### Claim 21 (depends on 20)

The apparatus of claim 20, said juxtaposed features comprising any of positive and negative topics, High CSAT vs. Low CSAT, Topic A vs. Topic B, Agent A vs. Agent B, or last week vs. this week.

### Claim 22 (depends on 20)

The apparatus of claim 20, said insights engine configured to:
access any one or more sources of textual data;
find and extract features/sentences that are the most discriminating between the sources of textual data;
apply feature importance measures to the extracted features to effect discriminatory feature selection;
produce a representation of the feature importance measures;
develop additional feature importance measures based upon relative incidence in positive/negative interactions;
leverage discriminatory power of multiple importance measures to provide an ensemble/voting approach to determining customer sentiment;
wherein said discriminatory features are determined as an output of a weighted averaging, and wherein the weights are determined over time by users voting on efficacy of insights that come with the use of different weights.

### Claim 23 (depends on 1)

The apparatus of claim 1, said processor configured for:
mining social media to drive better customer care via new issue identification, wherein said social media is used as a leading indicator of said new issues.

### Claim 24 (depends on 23)

The apparatus of claim 23, said processor configured for:
making recommendations for product purchase where new products get introduced continuously and where monthly/quarterly product level training provided is not enough to keep agents knowledgeable.

### Claim 25 (depends on 1)

The apparatus of claim 1, said processor configured for:
leveraging social media to improve performance of chat and other customer care channels.

### Claim 26 (depends on 1)

The apparatus of claim 1, said processor configured for:
extracting parts of speech information for each feature in each selected topic; and
assigning vertical based weightages for point of sale (POS) tags;
wherein some POS tags are weighted higher than others; and
wherein some parts of speech are more likely to identify an issue than others.

### Claim 27 (depends on 1)

The apparatus of claim 1, said processor configured for:
applying text mining to new topics to identify leading topics and tail topics.

### Claim 28 (depends on 1)

The apparatus of claim 1, said processor configured for:
making recommendations based upon structured and unstructured reviews to support customer service agents.

### Claim 29 (depends on 28)

The apparatus of claim 28, said processor configured for:
providing a product recommendation engine summarizing product details from structured reviews and unstructured user reviews;
extracting product details from unstructured reviews; and
using said product recommendation engine to equip agents for making sales pitches to customers.

### Claim 30 (depends on 29)

The apparatus of claim 29, said product recommendation engine configured for:
providing a summarization and sentiment analysis engine for leveraging summarization and sentiment analysis.

### Claim 31 (depends on 29)

The apparatus of claim 29, said processor configured for:
using a link to quantify a context relationship between a product and a feature.

### Claim 32 (depends on 29)

The apparatus of claim 29, said processor configured for:
from a set of collected reviews for a given product, developing a set of keywords/features which comprise possible attributes for a given product;
identifying if said set of keywords/features comprises attributes by using an attribute extraction engine to semantically determine if a given keyword/feature is an attribute or not;
using different parameters to determine a relationship between a possible attribute and a given product;
searching to determine a search co-occurrence strength measure based co-occurrence on a results/hits count;
using said parameters as input features for a model to determine if a given word is an attribute;
once prospective attributes are identified, inputting lines consisting of said keywords/features for said summarization and sentiment analysis engine;
performing sentiment analysis at the attribute level to determine individual attributes' sentiment; and
picking up lines from existing texts that contain key attributes and the sentiment score.

### Claim 33 (depends on 32)

The apparatus of claim 32, said processor configured for:
defining a weighting function/score for a combination of a frequency of occurrence of an attribute with a product and a sentiment associated with said attribute;
posting attributes with a highest score and a high scoring sentences associated with said highest scoring attributes as recommendations;
wherein said score identifies attribute characteristics that are most representative of a product; and
wherein said score is associated with any of very good or very poor sentiments.

### Claim 34 (depends on 1)

The apparatus of claim 1, said processor further configured for:
selecting posts that are not covered by an a priori topic model;
selecting rare terms occurring for a first time in a period compared to a recent past;
calculating term frequency inverse document frequency (TFIDF) scores for the rare terms;
tagging each rare term as a part of speech;
applying vertical based weightage of point of sale (POS) tags;
calculating a score of each post as a summation of weighted TFIDF scores of the rare terms; and
selecting top posts.