Predicting and visualizing outcomes using a time-aware recurrent neural network
Disclosed systems and methods predict and visualize outcomes based on past events. For example, an analysis application encodes a sequence of events into a feature vector that includes, for each event, a numerical representation of a respective category and a respective timestamp. The application applies a time-aware recurrent neural network to the feature vector, resulting in one or more of (i) a set of future events in which each event is associated with a probability and a predicted duration and (ii) a sequence embedding that contains information about predicted outcomes and temporal patterns observed in the sequence of events. The application applies a support vector model classifier to the sequence embedding. The support vector model classifier computes a likelihood of a categorical outcome for each of the events in the probability distribution. The application modifies interactive content according to the categorical outcomes and probability distribution.
1. A method comprising:
accessing a sequence of events, wherein each event comprises a timestamp at which the event occurred and a category from a plurality of predefined categories;
encoding the sequence of events into a feature vector comprising, for each event, a numerical representation of (i) a respective category and (ii) a respective timestamp;
applying, to the feature vector, a time-aware recurrent neural network that outputs a sequence embedding comprising a probability distribution of a plurality of future events and an associated duration for each future event, wherein the time-aware recurrent neural network accommodates time irregularities in the sequence of events;
applying, to the sequence embedding, a support vector model classifier that computes a likelihood of a categorical outcome for each of the events in the probability distribution; and
providing one or more of (i) the probability distribution or (ii) the categorical outcome to a user interface.
2. The method of claim 1 , further comprising:
accessing a plurality of pairs of training data, wherein each pair of training data comprises a sequence of events, an expected next event, and an expected categorical outcome;
training the time-aware recurrent neural network by iteratively:
obtaining a next event by applying the time-aware recurrent neural network to each pair of training data;
adjusting one or more parameters of the time-aware recurrent neural network such that a probability of the next event matching the expected next event is maximized; and
extracting a sequence embedding from the time-aware recurrent neural network; and
training the support vector model classifier with each of the extracted sequence embeddings by iteratively:
obtaining a categorical outcome by applying the support vector model classifier to an extracted sequence embedding; and
adjusting one or more parameters of the support vector model classifier such that a probability of the categorical outcome matching the expected categorical outcome is maximized.
3. The method of claim 1 , further comprising:
updating, via the user interface, one or more of (i) a timestamp or (ii) a category of an event in the sequence of events;
generating an updated feature vector based on the updated sequence;
computing an updated probability distribution; and
updating the user interface with the updated probability distribution.
4. The method of claim 1 , further comprising:
updating, via the user interface, the sequence of events to include an additional future event;
generating an updated feature vector based on the updated sequence;
computing an updated probability distribution; and
updating the user interface with the updated probability distribution.
5. The method of claim 1 , further comprising:
determining, from the plurality of future events, a sequence of future events comprising a first event and a second event;
determining that a cumulative probability of the first event and the second event occurring is greater than a cumulative probability of other sequences within the plurality of future events; and
presenting the sequence of events and associated aggregated outcomes on the user interface.
6. The method of claim 1 , wherein the sequence of events is determined based on interactions between a user device and an interactive computing environment.
7. The method of claim 1 , wherein the sequence of events comprises events derived from one or more activities of a plurality of users and wherein the probability distribution of a plurality of future events and the associated durations for each event are based on the activities of the plurality of users.
8. The method of claim 1 , further comprising:
presenting, to the user interface, a predefined list of categories;
receiving, from the user interface, a selection of a category from the predefined list of categories;
determining one or more additional events, that if added to the sequence of events, would result in a categorical outcome matching the category; and
presenting the one or more additional events to the user interface.
9. A system comprising:
a processing device; and
a memory device communicatively coupled to the processing device and comprising program instructions, wherein when executed, cause the processing device to perform operations comprising:
accessing a sequence of events, wherein each event comprises a timestamp at which the event occurred and a category from a plurality of predefined categories;
encoding the sequence of events into a feature vector comprising, for each event, a numerical representation of (i) a respective category and (ii) a respective timestamp;
applying, to the feature vector, a time-aware recurrent neural network that outputs a sequence embedding comprising a probability distribution of a plurality of future events and an associated duration for each future event, wherein the time-aware recurrent neural network accommodates time irregularities in the sequence of events;
applying, to the sequence embedding, a support vector model classifier that computes a likelihood of a categorical outcome for each of the events in the probability distribution; and
providing one or more of (i) the probability distribution or (ii) the categorical outcome to a user interface.
10. The system of claim 9 , wherein the operations further comprise:
accessing a plurality of pairs of training data, wherein each pair of training data comprises a sequence of events, an expected next event, and an expected categorical outcome;
training the time-aware recurrent neural network by iteratively:
obtaining a next event by applying the time-aware recurrent neural network to each pair of training data;
adjusting one or more parameters of the time-aware recurrent neural network such that a probability of the next event matching the expected next event is maximized; and
extracting a sequence embedding from the time-aware recurrent neural network; and
training the support vector model classifier with each of the extracted sequence embeddings by iteratively:
obtaining a categorical outcome by applying the support vector model classifier to an extracted sequence embedding; and
adjusting one or more parameters of the support vector model classifier such that a probability of the categorical outcome matching the expected categorical outcome is maximized.
11. The system of claim 9 , wherein the operations further comprise:
updating, via the user interface, one or more of (i) a timestamp or (ii) a category of an event in the sequence of events;
generating an updated feature vector based on the updated sequence;
computing an updated probability distribution; and
updating the user interface with the updated probability distribution.
12. The system of claim 9 , wherein the operations further comprise:
updating, via the user interface, the sequence of events to include an additional future event;
generating an updated feature vector based on the updated sequence;
computing an updated probability distribution; and
updating the user interface with the updated probability distribution.
13. The system of claim 9 , wherein the operations further comprise:
determining, from the plurality of future events, a sequence of future events comprising a first event and a second event;
determining that a cumulative probability of the first event and the second event occurring is greater than a cumulative probability of other sequences within the plurality of future events; and
presenting the sequence of events and associated aggregated outcomes on the user interface.
14. The system of claim 9 , wherein the sequence of events is determined based on interactions between a user device and an interactive computing environment.
15. A non-transitory computer-readable medium storing instructions that, when executed by one or more processing devices, cause the one or more processing devices to perform operations comprising:
accessing a sequence of events, wherein each event comprises a timestamp at which the event occurred and a category from a plurality of predefined categories;
encoding the sequence of events into a feature vector comprising, for each event, a numerical representation of (i) a respective category and (ii) a respective timestamp;
applying, to the feature vector, a time-aware recurrent neural network that outputs a sequence embedding comprising a probability distribution of a plurality of future events and an associated duration for each future event, wherein the time-aware recurrent neural network accommodates time irregularities in the sequence of events;
applying, to the sequence embedding, a support vector model classifier that computes a likelihood of a categorical outcome for each of the events in the probability distribution; and
providing one or more of (i) the probability distribution or (ii) the categorical outcome to a user interface.
16. The computer-readable medium of claim 15 , wherein the operations further comprise:
accessing a plurality of pairs of training data, wherein each pair of training data comprises a sequence of events, an expected next event, and an expected categorical outcome;
training the time-aware recurrent neural network by iteratively:
obtaining a next event by applying the time-aware recurrent neural network to each pair of training data;
adjusting one or more parameters of the time-aware recurrent neural network such that a probability of the next event matching the expected next event is maximized; and
extracting a sequence embedding from the time-aware recurrent neural network; and
training the support vector model classifier with each of the extracted sequence embeddings by iteratively:
obtaining a categorical outcome by applying the support vector model classifier to an extracted sequence embedding; and
adjusting one or more parameters of the support vector model classifier such that a probability of the categorical outcome matching the expected categorical outcome is maximized.
17. The computer-readable medium of claim 15 , wherein the operations further comprise:
updating, via the user interface, one or more of (i) a timestamp or (ii) a category of an event in the sequence of events;
generating an updated feature vector based on the updated sequence;
computing an updated probability distribution; and
updating the user interface with the updated probability distribution.
18. The computer-readable medium of claim 15 , wherein the operations further comprise:
updating, via the user interface, the sequence of events to include an additional future event;
generating an updated feature vector based on the updated sequence;
computing an updated probability distribution; and
updating the user interface with the updated probability distribution.
19. The computer-readable medium of claim 15 , wherein the operations further comprise:
determining, from the plurality of future events, a sequence of future events comprising a first event and a second event;
determining that a cumulative probability of the first event and the second event occurring is greater than a cumulative probability of other sequences within the plurality of future events; and
presenting the sequence of events and associated aggregated outcomes on the user interface.
20. The computer-readable medium of claim 15 , wherein the sequence of events is determined based on interactions between a user device and an interactive computing environment.