International Business Machines Corporation

Tao-Hung Jung

Ching-Chun Liu

Ci-Wei Lan

Cheng-Fang Lin

Yu-Siang Chen

Po-cheng Chiu

I-Chien Lin

Jui Hsiang Huang

Norman Kung

Cheng-Lun Yang

Tzu-Chen Chao

Yu-siang Chen

In an approach to detecting the transmission of messages, analyzing said messages, calculating a message risk score and transmitting a warning notification, one or more computer processors detect transmission of a message from a user to a selected recipient. The one or more computer processors extract message information from the detected message. The one or more computer processors retrieve one or more historical conversations between the user and the selected recipient of the detected message. The one or more computer processors determine a risk score corresponding to sending the detected message to the selected recipient based on applying the extracted message information and the retrieved historical conversations to a cognitive model.

Mistaken message prevention based on multiple classification layers

Embodiments of the present invention provide a method, system and computer program product for telephonic fraud prevention. In an embodiment of the invention, a method of telephonic fraud prevention includes speech recognizing in real-time, audio of a telephone call from a caller to a called party. Once the real-time audio has been speech recognized into text, the text may be parsed into words and the words compared to a hierarchical semantic tree of terms. On the condition that a match is detected between one of the words and a node of the tree, a level of the tree is identified at which the match is detected and the level is correlated to a degree of risk of the telephone call. Finally, an alert is displayed to the called party reflecting the degree of risk.

Cognitive telephone fraud detection

Aspects generate recommendations in response to messaging content, wherein processors are configured to identify keywords within text content extracted from a received message that are assigned to a topic leaf node within an event taxonomy and a root event node from which the topic leaf node depends; identify a plurality of recommendation items that each match a type of event of the root event node and are linked to the topic node of the identified keyword; determine, for each of the recommendation items, a sender sentiment value as a function of a preference of the sender, and a recipient sentiment value as a function of a preference of the recipient; and present to the sender and the recipient as an event recommendation a subset of the recommendation items that each satisfy a sentiment threshold for the sender sentiment value and the recipient sentiment value.

Recommendation generator responsive to messaging content

Growing community of inventors

Taipei, Taiwan

Tao-Hung Jung