Systems and methods are provided for performing operations including: receiving, by a messaging application, input that selects an option to generate a message using an avatar with a caption; presenting, by the messaging application, the avatar and a caption entry region proximate to the avatar; populating, by the messaging application, the caption entry region with a text string comprising one or more words; determining, by the messaging application, context based on the one or more words in the text string; and modifying, by the messaging application, an expression of the avatar based on the determined context.

A set of unstructured data is analyzed to infer structural elements from the unstructured data, and quantized data quality levels, indicative of data quality in the structural elements, are assigned to the structural elements. A set of structured data is generated to include the structural elements inferred from the unstructured data and associations between respective ones of the structural elements in the set of structured data and the corresponding quantized quality levels assigned to the structural elements. The set of structured data, including the associations between respective ones of the structural elements and the corresponding quantized quality levels assigned to the structural elements, is provided to a user interface application to enable the user interface application to visually display varying data qualities in the set of structured data.

A system and method for classifying digital images is presented. The method includes extracting a plurality of descriptive data items of a transaction evidence from a digital image indicating a plurality of purchased items; searching in data source for informative data based on the extracted plurality of descriptive data items, wherein the informative data includes a price; determining a correlated amount for each of at least one of the plurality of descriptive data items, wherein the correlated amount determined for one of the descriptive data items defines a paid price for the descriptive data item; determining, based on at least one expense type classification rule, a primary expense type of the transaction evidence, wherein the at least one expense type classification rule is applied to the plurality of descriptive data items and each of the correlated amount; and classifying the digital image based on the primary expense type.

A system and method for classifying digital images is presented. The method includes extracting a plurality of descriptive data items of a transaction evidence from a digital image indicating a plurality of purchased items; searching in data source for informative data based on the extracted plurality of descriptive data items, wherein the informative data includes a price; determining a correlated amount for each of at least one of the plurality of descriptive data items, wherein the correlated amount determined for one of the descriptive data items defines a paid price for the descriptive data item; determining, based on at least one expense type classification rule, a primary expense type of the transaction evidence, wherein the at least one expense type classification rule is applied to the plurality of descriptive data items and each of the correlated amount; and classifying the digital image based on the primary expense type.

Some embodiments may perform operations of a process that includes obtaining a natural language text document and use a machine learning model to generate a set of attributes based on a set of machine-learning-model-generated classifications in the document. The process may include performing hierarchical data extraction operations to populate the attributes, where different machine learning models may be used in sequence. The process may include using a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model augmented with a pooling operation to determine a BERT output via a multi-channel transformer model to generate vectors on a per-sentence level or other per-text-section level. The process may include using a finer-grain model to extract quantitative or categorical values of interest, where the context of the per-sentence level may be retained for the finer-grain model.

Some embodiments may perform operations of a process that includes obtaining a natural language text document and use a machine learning model to generate a set of attributes based on a set of machine-learning-model-generated classifications in the document. The process may include performing hierarchical data extraction operations to populate the attributes, where different machine learning models may be used in sequence. The process may include using a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model augmented with a pooling operation to determine a BERT output via a multi-channel transformer model to generate vectors on a per-sentence level or other per-text-section level. The process may include using a finer-grain model to extract quantitative or categorical values of interest, where the context of the per-sentence level may be retained for the finer-grain model.

A hybrid entity recognition system and accompanying method identify composite entities based on machine learning. An input sentence is received and is preprocessed to remove extraneous information, perform spelling correction, and perform grammar correction to generate a cleaned input sentence. A POS tagger tags parts of speech of the cleaned input sentence. A rules based entity recognizer module identifies first level entities in the cleaned input sentence. The cleaned input sentence is converted and translated into numeric vectors. Basic and composite entities are extracted from the cleaned input sentence using the numeric vectors.

This disclosure enables various technologies that can (1) learn new synonyms for a given concept without manual curation techniques, (2) relate (e.g., map) some, many, most, or all raw named entity recognition outputs (e.g., “United States”, “United States of America”) to ontological concepts (e.g., ISO-3166 country code: “USA”), (3) account for false positives from a prior named entity recognition process, or (4) aggregate some, many, most, or all named entity recognition results from machine learning or rules based approaches to provide a best of breed hybrid approach (e.g., synergistic effect).

This disclosure enables various technologies that can (1) learn new synonyms for a given concept without manual curation techniques, (2) relate (e.g., map) some, many, most, or all raw named entity recognition outputs (e.g., “United States”, “United States of America”) to ontological concepts (e.g., ISO-3166 country code: “USA”), (3) account for false positives from a prior named entity recognition process, or (4) aggregate some, many, most, or all named entity recognition results from machine learning or rules based approaches to provide a best of breed hybrid approach (e.g., synergistic effect).

A method for remote damage assessment of a vehicle is provided. The present disclosure relates to the technical field of artificial intelligence, in particular to the technical field of image and text recognition. An implementation solution is: performing data collection on a target vehicle to determine damage information of the target vehicle; obtaining call content of an insurance claiming call for the target vehicle, and extracting accident-related information from the call content, wherein the accident-related information includes named entities in the call content and a relationship between the named entities; and determining a first fraud probability corresponding to the target vehicle at least based on the damage information and the accident-related information.