Method and system for unstructured information analysis using a pipeline of ML algorithms
A system and a method for increasing the classification confidence, with lesser dependence on large sets of training data, obtained by one or more machine learning based algorithms, by analyzing unstructured information using unstructured analysis pipeline comprising a probabilistic network such as a Bayesian network. The probabilistic network may comprise nodes associated with elements and cues defined by experts, and require fewer labelled data samples to train. The confidence level of the elements may be determined by machine learning and unstructured analysis methods and processed by the probabilistic network to estimate the confidence for a characterization quantity.
1. A system for inferring a probability the system comprising:
a processor adapted to execute a code for:
receiving an unstructured data sample;
extracting at least one sample slice from the unstructured data sample;
inferring a plurality of confidence values, each assigned to an element representing an interim factor indicated to be informative about at least one cue of a plurality of cues, by applying at least one machine learning model on the at least one sample slice;
feeding the plurality of confidence values into a probabilistic network having a plurality of nodes wherein each of the plurality of nodes represents the element; wherein the probabilistic network is a Bayesian network wherein the node associated with the at least one characterization quantity is connected to a plurality of nodes, each associated with a cue from a plurality of cues, through edges representing the relations between the values and probabilities;
calculating a plurality of additional confidence values, each assigned to the at least one cue from the plurality of cues, indicated to be informative about the probability of presence of a characterization quantity, by processing the plurality of confidence values using an associated layer of the probabilistic network; and
inferring the probability of presence for the characterization quantity, by processing the plurality of additional confidence values using another associated layer of the probabilistic network.
2. The system of claim 1 , wherein the plurality of nodes associated with a cue, are connected to additional pluralities of nodes, each node from the additional pluralities of nodes associated with an element, through edges representing the relations between the values and probabilities.
3. The system of claim 1 , wherein a confidence measure for the at least one characterization quantity is a non-decreasing function of a plurality of additional confidence measures associated with the at least one characterization quantity.
4. The system of claim 1 , wherein a cue confidence measure from the additional plurality of confidence measures is a non-decreasing function of the plurality of confidence measures for the plurality of elements associated with the cue.
5. The system of claim 1 , wherein extracting the at least one sample slice comprises applying separators intrinsic to the unstructured data.
6. A computer implemented method for training at least one machine learning model for inferring a probability, comprising:
receiving an unstructured data sample;
when the unstructured data sample was received without an associated label, generating the associated label using an unstructured analysis algorithm;
extracting at least one sample slice from the unstructured data sample;
inferring a plurality of confidence values, each assigned to an element representing an interim factor indicated to be informative about at least one cue of a plurality of cues, by applying the at least one machine learning model on the at least one sample slice;
feeding the plurality of confidence values into a probabilistic network having a plurality of nodes wherein each of the plurality of nodes represents an element; wherein the probabilistic network is a Bayesian network wherein the node associated with the at least one characterization quantity is connected to a plurality of nodes, each associated with a cue from a plurality of cues, through edges representing the relations between the values and probabilities;
calculating a plurality of additional confidence values, each assigned at least one cue of the plurality of cues, indicated to be informative about the probability of presence of a characterization quantity, by processing the plurality of confidence values using an associated layer of the probabilistic network;
inferring the probability of presence for the characterization quantity, by processing the plurality of additional confidence values using another associated layer of the probabilistic network;
updating parameters of the probabilistic network representing the indicated prior of the co-occurrence characteristics, so that the probability of presence for the characterization quantity inferred is increased when the characterization quantity complies with the associated label; and
updating parameters of the probabilistic network representing an indicated prior of a co-occurrence characteristics, so that the probability of presence for the characterization quantity inferred is decreased when the characterization quantity does not comply with the associated label.
7. The computer implemented method of claim 6 , further comprising using a plurality of sample slices extracted from additional samples of unstructured data.
8. The computer implemented method of claim 7 , further comprising assigning labels to the plurality of sample slices.
9. The computer implemented method of claim 8 , further comprising using models from the at least one machine learning model on the labelled sample slices and estimating a plurality of back-propagated element confidence values from associated cues, associated with the at least one characterization quantity.
10. The computer implemented method of claim 9 , further comprising estimating the confidence measure for the at least one characterization quantity using the probabilistic network, and the plurality of back-propagated element confidence values.
11. A computer implemented method, using a probabilistic network and a plurality of machine learning models, the computer implemented method comprising:
receiving at least one characterization quantity, a plurality of cues associated with the at least one characterization quantity, and at least one cue of the plurality of cues having a plurality of elements associated therewith;
receiving at least one sample of unstructured data;
extracting at least one sample slice from the sample of unstructured data;
generating a plurality of confidence measures for a plurality of elements by processing the at least one sample slice using at least one of the plurality of machine learning models;
generating an additional plurality of confidence measures for the plurality of cues based on confidence measures of the plurality of elements using the probabilistic network; wherein the probabilistic network is a Bayesian network wherein the node associated with the at least one characterization quantity is connected to a plurality of nodes, each associated with a cue from the plurality of cues, through edges representing the relations between the values and probabilities; and
generating a confidence measure for the at least one characterization quantity, based on confidence measures of the at least one of the plurality of cues using the probabilistic network.
12. The computer implemented method of claim 11 , wherein the plurality of nodes associated with a cue, are connected to additional pluralities of nodes, each node from the additional pluralities of nodes associated with an element, through edges representing the relations between the values and probabilities.
13. The computer implemented method of claim 11 , wherein the confidence measure for the at least one characterization quantity is a non-decreasing function of the plurality of additional of confidence measures associated with cues associated with the at least one characterization quantity.
14. The computer implemented method of claim 11 , wherein a cue confidence measure from the additional plurality of confidence measures is a non-decreasing function of the plurality of confidence measures for the plurality of elements associated with the cue.
15. The computer implemented method of claim 11 , wherein extracting the at least one sample slice comprises applying separators intrinsic to the unstructured data.
16. The computer implemented method of claim 11 , further comprising:
updating parameters of the probabilistic network representing an indicated prior to a co-occurrence characteristic, so that the probability of presence for the characterization quantity inferred is increased when the characterization quantity complies with the associated label.
17. The computer implemented method of claim 11 , further comprising:
updating parameters of the probabilistic network representing an indicated prior to a co-occurrence characteristic, so that the probability of presence for the characterization quantity inferred is decreased when the characterization quantity does not comply with the associated label.