Managing data in a cache memory of a question answering system

Patent metadata

Field	Value
Publication number	US-11200281-B2
Application number	US-201615079925-A
Country	US
Kind code	B2
Filing date	Mar 24, 2016
Priority date	Oct 7, 2014
Publication date	Dec 14, 2021
Grant date	Dec 14, 2021

Abstract

A technique for caching evidence for answering questions in a cache memory of a data processing system (that is configured to answer questions) includes receiving a first question. The first question is analyzed to identify a first set of characteristics of the first question. A first set of evidence for answering the first question is loaded into the cache memory. A second question is received. The second question is analyzed to identify a second set of characteristics of the second question. A portion of the first set of evidence, whose expected usage in answering the second question is below a determined threshold, is unloaded from the cache memory.

First claim

What is claimed is:

1. A method for caching evidence for answering questions in a cache memory of a question answering system that is configured to answer questions, comprising:
receiving, by the question answering system, a first question;
analyzing, by the question answering system, the first question to identify a first set of characteristics of the first question;
loading based on a loading rate (LR) metric that defines an amount of data loaded per unit of time, by the question answering system, a first set of evidence into the cache memory of the question answering system for answering the first question, wherein a value of the LR metric is a function of one or more of a question type, an evidence size, and a question key characteristic and how much of the cache memory is required by new evidence that is to be loaded;
receiving, by the question answering system, a second question;
analyzing, by the question answering system, the second question to identify a second set of characteristics of the second question;
comparing, by the question answering system, the first set of characteristics to the second set of characteristics to determine an expected usage for the first set of evidence in answering the second question; and
unloading based on an unloading rate (UR) metric that defines an amount of data unloaded per unit of time from the cache memory, by the question answering system, a portion of the first set of evidence based on the expected usage which improves the ability of the question answering system in maintaining relevant evidence in the cache memory and thereby improves operation of the question answering system, wherein a value of the UR metric is lowered when current data loaded in the cache memory can be reused in answering a subsequent question and is increased when the current data loaded in the cache memory cannot be reused in answering the subsequent question, and wherein the value of the UR metric is different than a value of the LR metric and is a function of one or more of the question type, the evidence size, and the question key characteristic.

2. The method of claim 1 , wherein the unloading, by the question answering system, a portion of the first set of evidence is performed at a rate that is based on a relevance of the first set of characteristics to the second set of characteristics.

3. The method of claim 1 , wherein the first set of characteristics and the second set of characteristics each include a question type, an evidence size, primary search attributes, and an evidence score that are respectively associated with the first and second questions.

4. The method of claim 1 , further comprising:
loading additional evidence for a hypothesis related to one of the first and second questions into the cache memory when another question is not received.

5. The method of claim 1 , wherein the cache memory is shared between processors of a node and the question answering system is a computing system that includes multiple nodes.

6. The method of claim 1 , wherein the unloading, by the question answering system, a portion of the first set of evidence is performed at a rate that is based on a relevance of the first set of characteristics to subsequent question characteristics.

Assignees

• IBM

Inventors

• Allen Corville O
• Carter Bernadette A
• Ghosh Rahul
• Kozhaya Joseph N

Classifications

• G06F12/121 Primary

  using replacement algorithms · CPC title

• G06F16/172

  Caching, prefetching or hoarding of files · CPC title

• G06F12/0811

  with multilevel cache hierarchies · CPC title

• G06F16/3329

  Natural language query formulation · CPC title

• G06F16/3344

  using natural language analysis · CPC title