Unusual score generators for a neuro-linguistic behavioral recognition system

1. A processor-implemented method, comprising: generating, via one or more processors, anomaly scores for a neuro-linguistic model of sensor input data obtained from one or more sources, the generating including:

receiving a stream of symbols generated from an ordered stream of normalized vectors generated from sensor input data received from one or more sensor devices during a first time period;

generating a set of words based on an occurrence of groups of symbols from the stream of symbols;

determining a number of previous occurrences of a first single word of the set of words;

determining a number of previous occurrences of each additional word of a same length as the first single word;

determining a first anomaly score based on the number of previous occurrences of the first single word and the number of previous occurrences of words of the same length as the first single word;

determining anomaly scores for each word from the set of words based on a length of that word;

determining a maximum anomaly score based on a maximum of the first anomaly score and the anomaly scores; and

outputting a data structure including a selected word from the set of words, an alert directive based on the maximum anomaly score, and the anomaly score for the selected word.

2. The processor-implemented method of claim 1 , further comprising:

generating a syntax comprising at least one of the one or more word combinations and describing relationships between the words of the syntax;

determining the distance between the generated syntax and a syntax model, wherein the syntax and the syntax model comprise a connected graph, wherein each node in the connected graph represents one of the words in the stream, and wherein edges connecting the nodes represent probabilistic relationships between words in the stream; and

outputting a second anomaly score based on the determined distance.

3. The processor-implemented method of claim 2 , wherein determining the distance comprises:

determining a set of most significant words of the generated syntax based on the probabilistic relationships; and

comparing each word from the set of most significant words to words in the syntax model in to determine a distance between each word from the set of most significant words and words in the syntax model.

4. The processor-implemented method of claim 3 , wherein the comparing comprises:

determining a feature weight for features for each word from the set of most significant words and comparing the features and feature weight of each word to the words in the syntax model, wherein the feature weight is based on a summation and maximum of feature scores.

5. The processor-implemented method of claim 2 , wherein the distance between the generated syntax and the syntax model is weighted based on lengths of words in the generated syntax.

6. A computer-readable non-transitory storage medium storing instructions, that when executed on one or more processors, perform an operation for generating anomaly scores for a neuro-linguistic model of input data obtained from one or more sources, the instructions comprising instruction to:

receive a stream of symbols generated from an ordered stream of normalized vectors generated from sensor input data received from one or more sensor devices during a first time period;

generate a set of words based on an occurrence of groups of symbols from the stream of symbols;

determine a number of previous occurrences of a first single word from the set of words;

determine a number of previous occurrences of each additional word of a same length as the first single word;

determine a first anomaly score based on the number of previous occurrences of the first single word and the number of previous occurrences of words of the same length as the first single word;

determine anomaly scores for each word from the set of words based on a length of that word;

determine a maximum anomaly score based on a maximum of the first anomaly score and the anomaly scores; and

output a data structure including a selected word from the set of words, an alert directive based on the maximum anomaly score, and the anomaly score for the selected word.

7. The computer-readable non-transitory storage medium of claim 6 , further comprising:

generating a syntax comprising at least one of the one or more word combinations and describing the relationship between the words of the syntax;

determining the distance between the generated syntax and a syntax model, wherein the syntax and syntax model comprise a connected graph, wherein each node in the connected graph represents one of the words in the stream, and wherein edges connecting the nodes represent a probabilistic relationship between words in the stream; and

outputting a second anomaly score based on the determined distance.

8. The computer-readable non-transitory storage medium of claim 7 , wherein determining the distance comprises:

determining a set of most significant words of the generated syntax based on probabilistic relationships; and

comparing each word from the set of most significant words to words in the syntax model to determine a distance between each word from the set of most significant words and words in the syntax model.

9. The computer-readable non-transitory storage medium of claim 8 , wherein the comparing comprises:

determining a feature weight for features for each word from the set of most significant words and comparing the features and feature weight of each word to the words in the syntax model, the feature weight being based on a summation and maximum of feature scores.

10. The computer-readable non-transitory storage medium of claim 7 , wherein the distance between the generated syntax and a syntax model is weighted based on lengths of words in the generated syntax.

11. The processor-implemented method of claim 1 , wherein the determining the anomaly scores for each word from the set of words includes assigning a maximum anomaly score to a word from the set of words, in response to determining that the word does not match a word model.

12. The processor-implemented method of claim 1 , wherein the generating the set of words is based on a level-based learning model in which one-letter words are learned at a first level and two-letter words are learned at a second level.