Browse Prior Art Database

Answering Predictive Questions using Prediction Accuracy Profiles

IP.com Disclosure Number: IPCOM000237220D
Publication Date: 2014-Jun-09
Document File: 3 page(s) / 114K

Publishing Venue

The IP.com Prior Art Database

Abstract

A system and method for answering predictive questions using prediction accuracy profiles is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 39% of the total text.

Page 01 of 3

Answering Predictive Questions using Prediction Accuracy Profiles

Disclosed is a system and method for answering predictive questions using prediction accuracy profiles.

The system and method utilizes three main steps:
Gathering predictions - identifying predictions in unstructured text and identify the topic areas of the prediction


1.

Creating author profiles


2.

Determining if a given prediction became true/false


1.

Determining the accuracy of a given individual for a given topic

2.

Scoring evidence - answering questions about the future by finding other people's predictions and weighing them based on past prediction


3.

accuracy

Financial Example

® beat their earnings estimate this quarter?"

To answer this question, the system would crawl for all of the predictions of IBM's earning for this quarter. After it found the predictions, it would request the author profiles (step 2) of the people who wrote the predictions. Finally, knowing the past history of the authors, it would score the current predictions/evidence accordingly and provide an answer (step 3).

The system does the following:

Builds a predictive author profile from any prediction an author publishes


1.

Uses the predictive author profiles to answer prediction questions


2.

Deals with contradictory predictions using past author's prediction performance


3.

Author Profile

There are two components of the author profile. The first component is the author's predictions and the second component the author's accuracy of predictions for given topics. These two components are used in the process by:


(1) Gathering predictions to produce the author prediction profile
(2) Creating author profiles which consumes the prediction author profile and produces the author topic accuracy profile
(3) Scoring evidence which consumes the author topic accuracy profile

The author prediction profile has the authors associated with predictions and those predictions associated with topics.

The author topic accuracy profile contains at a minimum:

Author


Topic
Number of Correct Predictions
Number of Incorrect Predictions


(1) Gathering predictions

There are a couple steps that are required to gather the pertinent predictions:

  1A - Identify topics/categories of interest
1B - Crawl and index content to obtain a large corpus of statements (prediction candidates)
1C - For each prediction and author, label the prediction with one or more topic categories and add it to the author profile
The ordering of steps 1A, 1B, and 1C is not important. For example, a crawl of all prediction sources could be done before any question is asked to build up the author profiles. Alternatively, questions can be asked first to determine the topics necessary to build up a pertinent author profile.

1A - Identify topics/categories of interest


In one embodiment, the topic categories of interest are input to the system (step 1). Each topic is defined through either a list of hot words/phrases and/or through machine learning (training the s...